diff --git a/.coverage b/.coverage new file mode 100644 index 00000000..a6d89bc8 --- /dev/null +++ b/.coverage @@ -0,0 +1 @@ +!coverage.py: This is a private format, don't read it directly!{"lines":{"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/__init__.py":[12,14,15,18,19,20,22,23,24,26,27,29,30,31,32,33,34,35,37,38,39,41,42,43,45,46,47,48,50,51,52,53,55,56,57,58,59,60,62,64,66,68,69,70,71,72],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/__init__.py":[6,9,10,11,12,13,14,15,16,17,18,21,22,23,24,25,26,27],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/embedding.py":[128,129,130,131,4,133,7,8,11,12,13,140,15,141,142,18,146,148,143,144,145,155,157,39,41,169,43,45,174,47,48,177,178,49,51,181,182,52,55,185,186,179,60,61,63,193,68,199,72,201,73,75,76,205,82,85,86,87,89,90,91,93,104,111,119,120,121,122,123,124,125,126,127],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/utils.py":[4,5,6,7,9,42,43,12,44,45,46,16,24,57,26,27,28,25,31],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/__init__.py":[13,15,17,19,20,21,22,24,26,27,28,30,32,33,35,36,37,38,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,56,57,58,59,60,61,63,64,65,67,68,69,70,72,73,74,75,78,79,80,83,84,85,86,87,88,89,90,91,92,93,94],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/_logger.py":[1,130,131,4,132,134,7,8,9,10,11,137,13,140,15,16,143,19,20,24,25,26,155,27,29,30,31,32,33,45,46,47,49,50,51,52,53,56,78,79,80,83,84,88,92,94,95,99,100,101,102,103,106,107,108,110,114,119,125,127],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/batch.py":[4,6,7,8,11,13,14,15,16,18,19,20,21,24,29,32,33,34,35,36,37,38,40,42,43,44,45,47,50,57,58,59,60,61,62,63,64,65,67,68,69,70,73,74,75,76,80,81,83,84,85,87,92,99,100,101,102,103,105,106,108,109,112,113,114,115,116,117,119,120,122,124,125,126,127,129,130,131,132,133,135,136,138,139,141,146,171,174,175,176,177,178,181,182,183,184,185,186,187,189,190,193,194,202,204,207,211,215,223,224,225,226,227,228,229,230,233],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/sampler.py":[3,5,6,7,8,134,135,11,140,13,137,16,149,150,151,24,153,26,155,156,158,160,34,162,163,164,166,165,40,167,42,170,43,46,52,54,55,58,186,187,188,190,191,192,193,68,70,71,72,73,75,83,84,86,87,89,90,91,92,93,94,96,97,98,100,102,103,104,105,106,107,108,109,110,112,113,114,115,117,120],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/dataset.py":[515,516,518,532,543,550,552,554,560,562,570,578,585,586,587,589,590,592,606,607,608,609,610,611,617,619,631,632,633,634,635,640,641,643,660,676,688,694,696,702,704,722,723,725,726,727,728,729,734,737,738,859,740,742,751,752,753,754,755,756,757,758,862,760,761,762,763,764,765,766,767,768,770,771,772,774,791,792,793,794,795,796,285,287,290,291,803,293,294,806,296,297,298,299,300,301,302,303,811,305,809,824,314,316,317,318,319,320,321,834,835,836,837,838,322,323,324,325,326,327,328,334,329,332,337,849,338,339,340,342,335,344,857,858,347,348,861,345,350,346,865,864,863,866,860,351,868,354,353,356,871,867,875,869,870,360,363,364,877,365,367,369,883,884,886,376,377,378,379,380,381,382,383,384,385,386,387,388,894,895,896,897,402,409,410,412,413,415,420,421,422,423,425,426,427,431,432,434,872,441,443,445,447,451,452,453,454,459,873,474,807,486,487,488,490,491,493,499,500,502,503,505,506,507,509],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/field.py":[4,7,8,9,12,13,14,15,16,18,19,21,22,535,25,26,27,28,29,30,33,34,35,36,37,38,41,43,44,557,46,559,560,47,562,48,52,53,54,563,56,57,58,59,60,564,62,63,64,568,570,67,68,572,70,71,72,65,74,578,76,580,78,590,80,591,585,83,592,85,593,87,594,89,595,596,597,598,599,95,96,97,98,99,100,613,101,614,102,615,609,616,617,618,104,106,108,622,624,113,114,115,116,629,117,118,120,119,122,130,131,132,133,134,135,136,137,138,651,139,653,140,141,142,146,147,148,149,150,663,152,659,661,157,158,159,160,162,165,677,167,169,681,682,683,685,686,175,687,177,178,688,180,181,182,183,184,690,691,187,692,693,190,694,192,697,200,201,202,205,206,207,209,211,212,214,220,221,222,226,45,236,242,244,252,254,255,256,257,259,261,278,565,566,567,298,569,571,318,573,574,575,339,576,577,579,359,581,582,379,584,586,626,398,419,695,428,429,430,431,432,433,434,435,436,437,438,439,441,443,444,445,446,447,448,450,451,452,453,454,455,456,458,459,460,465,482,484,485,487,490,491,610],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/utils.py":[3,5,517,6,7,518,10,11,12,13,14,521,16,17,18,19,20,524,22,23,527,528,26,27,535,29,536,540,541,542,543,35,544,545,547,546,40,548,551,552,553,554,46,550,49,563,564,53,565,567,568,569,59,60,574,62,63,64,67,522,592,599,609,615,530,118,119,120,121,122,124,125,126,127,641,129,130,132,131,134,647,648,649,650,651,652,135,139,140,656,142,144,147,659,145,146,662,663,664,151,666,667,152,669,670,153,672,673,674,163,676,165,678,679,168,681,682,643,685,644,645,192,709,217,218,219,220,222,736,738,227,739,740,226,229,232,233,230,148,231,745,234,235,149,236,237,238,239,240,244,245,241,242,243,246,247,248,249,250,251,252,253,254,255,256,259,260,263,154,271,273,274,156,277,157,280,158,642,288,289,159,291,292,293,294,295,296,297,298,161,301,316,333,334,335,337,338,339,340,341,342,343,344,345,346,347,348,349,350,351,352,353,354,355,356,357,358,359,360,361,364,388,389,390,391,392,393,396,397,405,411,413,416,417,421,430,433,436,437,438,439,440,290,445,449,451,452,454,456,457,458,460,463,465,466,469,470,471,475,476,477,478,479,480,485,496,497,498,499,500,501,502,503,505,506,507,508,509,510,511],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/instance.py":[58,5,37,39,7,11,46,47,48,52,53,55,56,24,26,59,28,30],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/const.py":[4,7,11,29,30,31,32,33,34,35,36,37,39,42,43,45,51,56,61,64,65,67,70,71,73,76,77,79],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/callback.py":[1024,513,1026,1030,1031,1036,529,531,1043,1044,1060,1071,562,51,53,1077,55,56,57,58,59,60,61,62,63,64,65,66,576,68,69,1092,583,72,73,74,76,78,80,81,84,85,87,88,89,91,92,603,606,97,612,106,108,621,109,623,110,113,111,118,120,123,125,128,130,133,135,648,138,140,143,145,660,148,150,153,155,159,161,674,164,166,681,169,171,683,685,686,175,687,177,688,179,692,693,183,696,701,189,191,703,705,706,708,197,710,199,721,722,210,212,723,726,724,728,729,730,220,733,222,741,229,231,743,745,746,748,237,749,239,750,751,752,753,756,245,754,247,758,761,759,252,765,254,766,767,768,770,771,260,773,262,774,775,776,778,779,780,781,782,783,777,785,786,275,787,788,789,791,790,273,794,283,795,796,797,287,799,289,800,801,802,805,293,295,303,818,820,821,310,311,312,313,822,315,316,823,318,830,824,321,322,826,836,828,827,831,832,833,841,329,331,332,333,334,839,336,337,842,851,339,340,341,852,855,348,349,350,863,351,353,864,357,870,871,361,875,365,369,881,373,377,889,890,381,385,389,902,393,907,397,912,401,405,410,411,922,929,420,428,945,946,437,961,964,455,968,457,459,461,462,463,468,469,471,472,473,987,479,482,504,489,491,1003,492,493,494,496,1009,497,1014,1016,506,1020],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/tester.py":[34,35,37,38,40,41,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,62,66,92,94,95,97,100,102,103,104,105,106,107,109,110,111,118,119,121,127,131,132,134,138,139,141,148,149,150,151,152,153,154,155,156,158,159,160,162,164,165,166,167,170,171,173,174,176,177,178,181,182,183,184,185,187,188,189,190,191,192,194,195,196,197,199,206,207,209,211,213,214,215,217,223,224,225,226,227,228],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/metrics.py":[4,6,7,8,9,12,13,15,16,18,19,20,21,22,23,24,25,26,29,117,119,120,121,122,124,133,137,138,141,151,156,158,165,166,179,180,181,182,183,185,186,187,188,192,193,194,195,200,208,209,213,215,230,231,235,236,239,240,241,242,246,247,249,250,253,254,255,256,257,258,259,262,263,264,265,267,270,271,272,274,277,278,279,280,281,282,284,285,286,287,288,290,292,295,305,307,309,311,313,314,316,329,330,332,336,340,341,343,345,346,347,348,350,354,355,356,357,359,360,362,369,370,371,372,373,376,386,388,389,390,391,392,393,394,395,396,398,399,400,401,402,406,437,468,477,479,480,481,482,483,484,485,486,488,489,491,492,493,496,504,505,506,507,508,509,510,511,512,514,515,516,520,561,564,566,568,570,573,574,575,576,577,578,579,580,581,582,586,587,588,589,590,592,593,595,597,598,599,601,609,612,616,620,622,623,624,625,633,634,635,636,637,638,640,641,643,644,646,647,648,649,651,652,653,655,657,658,659,660,661,662,663,664,665,666,667,668,669,670,671,672,673,674,675,676,677,678,679,681,686,687,688,689,690,691,692,694,695,696,697,699,700,702,704,712,713,714,716,719,726,727,728,729,730,732,733,734,736,738,742,743,747,750,759,760,761,762,763,766,776,777,778,779,780,781,784,799,802,804,806,808,810,811,813,814,816,818,819,820,821,823,825,827,836,837,838,839,841,842,845,846,850,851,852,853,855,856,857,858,859,862,863,864,865,867,868,870,873,875,876,878,879,880,881,883,884,885,887,888,891,893,895,897,900,901,903,905,907,909,910,911,912,914,916,918,919,920,921,923,929,932,933,935,936,937,939,940,942,944,945,946,947,949],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/vocabulary.py":[4,7,8,11,12,13,15,16,17,18,21,26,35,40,42,43,44,46,49,54,56,57,58,59,61,62,64,67,90,92,93,94,95,96,97,98,99,100,102,104,105,116,117,118,120,121,133,134,135,137,145,146,147,148,149,150,151,153,154,166,168,169,181,182,184,190,191,192,193,194,195,197,198,199,200,201,202,203,204,205,206,207,209,214,215,217,219,221,229,231,242,244,251,252,253,254,258,259,273,279,280,282,283,285,287,289,291,292,295,296,297,301,302,303,304,305,311,313,317,337,338,342,343,344,345,346,348,349,350,352,354,355,356,358,359,360,361,368,369,370,371,377,379,385,387,398,400,401,406,407,408,410,411,416,417,418,420,428,430,443,447,448,450,451,453,457,458,460,463,465,466],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/_parallel_utils.py":[1,97,3,5,7,8,9,10,11,76,104,14,105,107],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/losses.py":[4,6,8,9,11,12,13,14,17,18,20,21,23,24,25,26,27,28,29,30,31,34,37,39,40,41,43,52,55,62,63,76,77,78,79,80,82,83,84,85,89,90,91,92,102,110,112,113,114,115,119,120,122,123,125,126,127,128,129,130,131,134,135,136,137,139,141,142,143,145,148,149,150,151,152,153,155,156,157,158,160,162,163,165,168,188,190,192,193,194,195,198,201,222,224,225,226,227,228,229,230,232,233,234,235,236,239,240,241,242,243,245,246,249,259,261,262,263,264,265,267,268,271,280,282,283,284,285,286,288,289,292,303,305,306,307,308,309,310,312,313,316,323,325,326,327,329,331,332,333,334,335,336,337,338,339,340,341,343,345,347,353,356,357,358,359,360,361,366,374,377,386,387,395,410,432],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/optimizer.py":[4,6,7,8,9,135,138,12,13,14,15,18,151,24,26,27,156,29,30,32,35,41,43,47,48,51,54,61,68,70,71,72,73,75,76,78,80,83,90,92,93,95,96,98,99,101,103,106],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/trainer.py":[517,518,519,521,522,523,524,525,526,527,528,529,530,531,532,533,534,536,537,538,539,540,541,545,547,548,551,552,553,554,555,556,557,558,559,560,561,562,564,565,567,570,571,573,593,594,598,599,600,601,602,603,604,606,607,608,609,619,620,621,622,623,624,625,626,627,628,629,630,634,635,637,639,640,641,643,644,645,646,647,648,649,650,651,652,653,654,656,657,658,659,660,662,663,666,667,668,669,672,673,674,676,677,679,680,681,682,683,685,686,687,688,689,690,691,693,694,695,696,697,698,700,701,705,707,708,711,712,713,715,716,717,720,721,722,723,724,725,727,728,857,730,737,740,742,746,747,352,749,750,751,752,755,757,764,765,766,768,775,777,800,802,812,813,816,818,823,824,825,826,827,829,319,831,321,832,835,324,325,326,833,328,329,330,843,332,333,841,847,336,848,338,851,340,341,342,343,344,339,853,854,855,349,350,351,856,345,346,858,347,348,864,865,868,869,353,354,355,356,358,872,873,875,876,877,878,879,880,881,882,883,884,885,886,887,888,889,890,891,892,893,895,896,898,899,900,901,902,903,904,905,907,908,909,910,911,913,914,915,916,917,918,919,920,924,925,927,928,418,932,936,425,426,427,937,941,942,939,940,431,943,433,944,945,947,437,438,948,949,441,954,950,444,951,958,449,450,961,962,964,454,965,456,968,458,970,971,974,466,482,484,485,489,490,491,498,499,502,503,506,507,510,511],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/static_embedding.py":[4,7,9,11,12,13,14,16,17,18,19,20,21,22,25,66,69,70,71,72,75,76,77,78,79,83,84,91,92,119,121,122,123,124,127,128,130,133,134,135,136,140,141,142,143,144,146,147,148,150,151,153,154,155,156,158,164,165,166,167,168,169,171,179,181,182,186,188,202,204,205,207,209,210,226,227,229,230,231,232,233,237,238,239,240,241,242,243,244,245,246,247,248,249,250,252,254,257,258,259,260,261,262,269,270,271,272,275,277,279,283,284,285,286,287,288,290,292,299,300,301,302,303,304],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/__init__.py":[13,15,17,19,21,22,23,24,25,26,28,29,30,31,32,33,34,35,37,38,40,42,43,44,45,46,48,50,51,52,53,54,55,57,58,59,60,61,63,65,66,67,68,69,70,71,72,73,74,75,76,78,79,83,84,85,87,88],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/embed_loader.py":[4,6,7,10,11,12,14,16,17,20,22,23,24,25,34,39,41,44,45,46,63,64,66,67,68,69,70,71,72,73,75,76,77,78,80,81,82,83,84,86,88,90,91,92,93,100,101,102,103,104,105,106,107,108,109,111,112,114,116,117,118,133,134,135,136,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,157,166,168,169,170,171,173,174,175,176,177,178,180,181,182,183,184,185,187,188,190],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/data_bundle.py":[4,6,9,10,13,142,27,29,30,31,159,33,45,55,184,64,74,203,83,92,117],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/model_io.py":[32,3,5,6,9,42,12,17,19,53,22,55,62],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/__init__.py":[44,47,49,50,51,52,53,54,56,57,58,59,60,61,62,63,65,66,68,70,71,72,73,74,76,77,78,79,80,81,82,83],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/classification.py":[1,259,4,5,6,7,8,9,261,264,12,13,14,15,16,17,19,20,21,279,24,291,164,45,47,304,50,178,180,306,309,183,72,73,201,339,244,119,120],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/loader.py":[65,66,1,4,33,70,7,67,9,10,11,12,68,78,15,19,21,22,24,63],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/file_utils.py":[4,7,8,9,10,11,14,15,16,17,18,19,21,22,23,25,28,29,30,32,33,35,36,38,40,41,43,44,45,46,50,51,52,53,54,58,60,61,62,63,64,65,66,67,68,69,71,73,74,76,77,78,79,83,84,85,86,87,88,89,90,91,92,93,94,96,97,98,99,102,103,104,107,108,109,110,114,159,186,202,228,252,273,293,306,418,427,434,443],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/utils.py":[33,34,35,4,36,7,10,11,12,14,17,81],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/conll.py":[1,4,5,6,7,8,9,10,11,12,15,16,17,18,19,146,21,22,23,24,25,150,278,28,279,282,286,287,408,421,175,177,183,62,446,64,448,451,325,204,78,208,92,349,222,273,224,351,354,227,404,117,405,119,125],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/file_reader.py":[33,34,3,35,5,7,9,41,42,12,43,78,47,44,24,25,26,30],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/csv.py":[32,1,34,33,4,35,36,7,8,9,10,37,13,24,26,27,28,29,30],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/cws.py":[1,4,38,7,8,9,10,11,39,13,14,15,47,18,56],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/json.py":[1,4,38,7,8,9,10,13,25,27],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/loader/matching.py":[1,129,4,5,6,7,8,11,12,13,15,16,17,18,19,20,273,277,23,159,35,37,40,170,298,300,303,184,186,189,318,66,216,98,228,109,241,243,246,120,122],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/__init__.py":[9,11,13,15,16,17,18,19,21,22,23,24,25,26,28,29,30,31,32,33,34,35,36,37,38,39,42,43,44,46,47,48],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/classification.py":[1,4,5,6,7,8,134,264,11,392,13,15,16,17,18,19,20,21,22,408,24,410,28,414,32,34,37,172,52,182,315,320,449,195,197,70,201,333,335,339,89,218,247,228,104,106,119,249,382],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/pipe.py":[1,4,7,10,13,14,23],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/utils.py":[1,66,153,4,5,6,39,9,137,11,12,15,87,121,91],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/conll.py":[1,4,5,6,7,8,9,12,13,14,15,16,17,18,19,20,141,272,23,286,288,34,36,293,43,306,308,182,313,192,328,330,79,208,210,215,225,98,227,100,233,113,114],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/matching.py":[128,1,129,259,4,5,6,7,8,9,10,11,12,13,14,15,135,140,18,19,20,21,22,146,147,25,152,260,134,169,42,171,44,177,50,265,266,191,64,141,271,272,247,248,122,123,253,254],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/io/pipe/cws.py":[1,4,7,8,136,10,11,12,13,14,17,155,157,34,168,50,65,202,84,110,254],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/__init__.py":[18,22,23,25,27,29,31,33,34,35,37,38,39,40,42,44,45,46,47,49,52,53,54,55,56],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/decoder/__init__.py":[4,6,7,8,9,12,13,14,15],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/decoder/crf.py":[1,4,5,8,9,11,12,15,29,31,32,33,34,35,36,37,38,40,41,42,43,44,46,47,48,50,51,52,53,54,55,56,57,58,59,60,63,73,74,75,76,93,94,95,96,97,98,102,121,122,123,124,125,126,127,128,157,170,173,175,177,178,181,182,183,184,186,187,192,194,196,204,205,206,207,209,211,212,213,214,215,216,218,219,221,223,231,232,233,236,237,238,240,242,243,244,245,246,247,248,250,252,261,262,263,264,265,267,269,282,283,284,287,288,289,290,291,295,296,297,298,299,300,301,302,303,304,306,310,311,312,314,316,317,318,319,320,321,322,323,328,329],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/utils.py":[4,134,7,8,11,12,14,15,16,19,35,37,39,41,43,45,47,49,52,54,56,57,60,61,62,63,64,65,67,68,69,70,72,73,74,75,77,80,83,120],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/decoder/mlp.py":[1,4,7,8,10,13,44,46,47,48,49,50,51,52,53,55,57,60,61,62,64,65,71,72,73,75,76,79,86,88,93,94,95,96,98,99],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/decoder/utils.py":[1,4,6,9],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/__init__.py":[4,9,10,12,14,16,18,20,21,22,24,25,26,27,29,32,33,34,35,36,37,38,39,40],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/attention.py":[128,1,4,132,7,9,10,11,13,16,20,22,23,24,25,26,27,28,30,38,39,40,41,42,43,46,175,55,184,57,186,58,59,60,61,62,64,65,66,67,69,198,70,71,73,74,75,76,77,78,80,212,88,89,90,92,93,94,97,98,99,100,101,102,105,106,107,110,126],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/bert.py":[512,4,517,7,10,11,12,13,14,15,17,18,20,21,22,24,25,28,30,44,571,70,586,587,75,76,77,591,78,79,80,81,82,83,84,85,600,86,87,92,100,107,621,110,115,119,632,125,126,129,133,136,654,149,150,153,154,667,155,156,158,159,160,161,162,165,167,169,170,171,172,173,689,177,178,180,181,182,183,184,187,188,189,703,191,192,193,194,197,198,199,200,715,204,205,206,208,209,210,212,214,727,215,216,217,219,220,221,222,224,225,226,229,230,743,744,232,747,235,239,241,242,243,244,245,248,249,250,251,252,253,255,256,257,258,259,262,263,776,264,265,266,268,269,270,271,274,275,786,276,277,278,279,283,796,284,285,286,289,290,291,292,293,294,296,809,297,298,299,300,303,304,816,305,306,307,308,310,311,312,313,314,317,318,319,320,833,321,323,324,325,326,327,328,329,330,331,334,335,848,336,337,338,340,852,854,343,344,345,346,349,877,374,376,377,378,385,386,387,388,389,390,391,393,396,909,399,400,401,402,403,404,406,407,409,410,417,424,425,427,428,429,430,431,432,433,434,435,437,500,509,510],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/char_encoder.py":[1,4,5,7,8,10,14,25,27,28,29,30,32,34,36,41,43,45,47,48,49,50,52,54,55,57,58,61,68,70,77,78,80,81,82,83,84,85,87,92,93,94,95,96,98,99],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/conv_maxpool.py":[1,4,6,7,8,11,23,25,26,28,29,32,33,36,37,38,43,52,59,60,69,77,79,80,81,82,84,85,86],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/lstm.py":[4,7,10,11,12,15,30,33,34,35,36,37,38,40,41,42,44,45,46,47,49,51,61,62,65,66,67,68,69,72,73,74,75,76,77,82],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/pooling.py":[1,129,4,5,6,7,135,9,10,137,13,141,25,27,38,62,67,69,73,85,86,88,92,102,107,109,114],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/star_transformer.py":[3,6,9,10,11,12,15,32,34,35,36,38,40,41,42,43,44,45,46,48,49,53,63,65,67,68,69,71,72,76,77,78,79,80,81,82,83,85,87,89,91,94,95,96,99,100,101,102,104,107,109,111,112,114,116,117,118,119,120,121,122,123,124,125,126,127,129,130,132,134,137,138,140,141,142,143,144,146,149,151,153,154,156,158,159,160,161,162,163,164,165,166],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/transformer.py":[1,4,6,8,9,12,26,28,29,30,31,32,33,34,35,36,37,39,46,47,48,49,50,51,52,54,55,56,58,65,66,69,70,71,72,73],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/dropout.py":[1,4,7,10,14,16,17,18,19,20,24],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/variational_rnn.py":[3,6,7,8,11,12,13,15,16,25,28,31,33,34,35,36,37,38,40,52,53,54,55,56,58,59,60,61,62,63,64,66,67,69,70,73,74,75,76,77,79,80,81,82,83,84,85,86,87,88,89,96,97,98,99,102,120,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,146,147,148,149,150,151,152,153,155,163,164,165,166,167,168,169,170,172,173,175,176,177,178,179,181,182,183,184,185,186,187,188,190,191,192,193,194,195,196,197,198,199,200,201,202,203,204,205,206,207,210,212,213,215,216,218,219,221,224,239,241,242,243,245,246,249,264,266,270,274,289,291,295],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/elmo_embedding.py":[4,7,136,10,11,12,13,14,15,141,17,18,19,20,21,23,155,163,171,173,305,58,61,92,99,111,119],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/modules/encoder/_elmo.py":[514,3,515,5,7,263,9,10,11,12,264,14,528,17,409,410,309,56,65,453,327,328,85,98,493,239,240,251,510],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/contextual_embedding.py":[99,4,7,104,10,12,76,14,15,16,17,18,19,20,23,24,27],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/bert_embedding.py":[4,7,8,135,11,12,14,15,16,17,271,19,20,21,22,23,24,149,273,27,157,168,171,186,67,198,71,203,207,211,215,95,98,227,361,115,250],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/char_embedding.py":[4,7,8,11,12,13,14,16,17,18,19,20,21,22,25,57,61,62,64,65,67,68,70,71,72,85,87,88,89,91,92,93,94,95,98,99,101,104,106,108,109,110,111,113,120,121,122,123,124,125,127,128,129,130,131,132,133,134,135,136,137,138,142,143,145,161,168,169,170,172,173,174,175,177,180,211,216,217,219,221,222,224,225,226,239,241,242,243,245,246,247,248,249,252,253,255,258,260,261,263,264,265,267,274,275,276,277,278,279,281,282,283,284,285,286,289,290,291,292,297,299,301,318],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/embeddings/stack_embedding.py":[4,7,10,12,13,15,18,37,39,40,41,42,43,44,45,46,48,49,50,51,52,53,55,64,71,75,87,92,99,100,101,102,103,104],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/__init__.py":[32,33,34,9,11,13,14,16,18,19,20,21,23,24,27,28,30,31],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/base_model.py":[32,1,33,3,5,7,10,12,14,15,17,20,24,25,26,27,29,30],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/bert.py":[4,6,8,10,11,13,14,15,16,17,20,57,58,59,60,61,65,67,68,69,71,77,78,80,81,82,83,84,86,91,93,98,135,136,137,138,139,142,144,145,146,148,154,155,156,157,158,159,160,161,162,164,169,171,176,215,216,217,218,219,222,224,225,226,228,234,235,236,237,239,251,253,258,300,301,302,303,306,308,311,313,319,320,321,322,323,324,326,343,345],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/biaffine_parser.py":[3,5,517,6,520,9,10,11,12,522,14,523,16,17,18,19,20,21,22,23,24,25,530,536,28,539,542,534,544,33,34,35,36,37,38,39,40,41,545,546,547,548,46,47,48,49,50,51,52,53,45,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,73,74,75,76,524,77,78,79,80,525,81,84,82,87,526,527,92,93,94,95,528,96,97,99,101,102,103,104,105,107,108,109,110,111,112,531,114,115,116,117,118,119,120,121,122,532,124,125,126,533,128,131,136,138,139,141,142,151,152,153,154,155,156,157,158,160,161,170,171,172,173,174,175,176,177,178,179,182,188,190,191,192,193,194,195,198,200,549,207,208,209,210,211,42,214,43,222,44,224,225,226,227,229,236,237,238,241,262,275,276,277,278,279,280,281,282,283,284,285,286,287,288,289,290,291,292,293,294,295,296,297,305,306,307,308,310,311,312,313,314,315,316,317,318,322,323,324,325,326,327,328,329,330,331,333,334,335,336,337,338,339,341,342,344,362,366,368,369,371,372,373,376,377,378,379,380,381,382,383,385,386,387,391,392,393,394,397,400,402,403,405,406,416,417,418,419,420,421,422,424,437,438,439,440,441,442,443,444,445,446,447,449,450,451,452,453,454,456,469,470,471,472,473,474,477,489,493,494,495,496,497,498,499,502],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/cnn_text_classification.py":[4,7,10,11,13,14,15,16,19,32,38,39,42,43,44,45,46,47,48,50,57,58,59,60,62,63,64,65,67,74,75,76],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/sequence_labeling.py":[3,5,6,10,11,12,14,15,16,17,18,19,20,21,22,25,39,41,61,75,78,82,93,95,96,98,99,100,101,102,104,112,113,114,116,118,120,122,124,132,134,136,138,140,141,143,151,152,153,154,155,156,158,159,160,161,162,163,165,170,171,174,189,191,193,195,196,197,198,199,200,201,202,203,204,206,207,213,218,219,221,229,230,231,232,233,234,236,237,238,239,240,241,243,252,253,254,257,259,263,264,267,269,270,271,272,273,274,275,277,279,287,289,296],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/snli.py":[4,6,9,10,11,12,14,15,16,17,20,32,35,36,38,41,42,43,44,45,48,49,50,51,52,54,57,58,59,60,61,63,65,66,68,77,78,79,80,81,82,83,87,89,90,91,92,94,95,99,100,101,102,104,105,107,113,115,116,117,121,122,123,124,126,127,128,129,130,131,134,136,137,138,139,142,143,144,145,146,147,148,149,151,153,154,155,156,158,160,162,165,167,168,169,174,177,178,179,182,183,184,185,186,187,189,190,193,194,195,196,197,198,199,202,204,205,208,209,211,213,214,215,216,217,218,220],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/models/star_transformer.py":[3,5,6,7,8,11,12,14,15,16,17,20,36,38,46,47,48,49,51,52,53,54,55,56,58,67,68,69,70,73,74,75,76,77,78,79,80,83,84,85,88,89,90,91,92,93,94,95,96,99,100,101,102,105,123,133,134,135,136,137,138,139,140,141,142,143,145,152,153,154,155,156,158,165,166,167,170,188,198,199,200,201,202,203,204,205,206,207,208,210,217,218,219,220,221,223,230,231,232,235,253,263,264,265,266,267,268,269,270,271,272,273,275,284,285,287,288,289,291,292,293,294,296,305,306,307],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/dist_trainer.py":[3,4,5,6,7,9,10,11,12,13,14,15,16,18,19,20,21,22,23,24,25,26,152,29,30,157,34,169,47,304,50,179,183,312,58,320,332,343,355,229],"/hdd/fudanNLP/fastNLP/fastNLP/fastNLP/core/predictor.py":[1,4,7,9,11,12,13,14,17,25,27,28,31,32,33,35,42,44,47,48,49,50,51,53,56,58,59,60,61,62,64,67,68,69,70,80,81]}} \ No newline at end of file diff --git a/.travis.yml b/.travis.yml index 210d158a..bd7a34f5 100644 --- a/.travis.yml +++ b/.travis.yml @@ -8,7 +8,7 @@ install: - pip install pytest-cov # command to run tests script: - - pytest --cov=./ test/ + - pytest --cov=fastNLP test/ after_success: - bash <(curl -s https://codecov.io/bash) diff --git a/README.md b/README.md index 476c129f..531fbc83 100644 --- a/README.md +++ b/README.md @@ -6,11 +6,12 @@ ![Hex.pm](https://img.shields.io/hexpm/l/plug.svg) [![Documentation Status](https://readthedocs.org/projects/fastnlp/badge/?version=latest)](http://fastnlp.readthedocs.io/?badge=latest) -fastNLP 是一款轻量级的 NLP 处理套件。你既可以使用它快速地完成一个序列标注（[NER](reproduction/seqence_labelling/ner)、POS-Tagging等）、中文分词、[文本分类](reproduction/text_classification)、[Matching](reproduction/matching)、[指代消解](reproduction/coreference_resolution)、[摘要](reproduction/Summarization)等任务； 也可以使用它构建许多复杂的网络模型，进行科研。它具有如下的特性： +fastNLP 是一款轻量级的 NLP 工具包。你既可以使用它快速地完成一个序列标注（[NER](reproduction/seqence_labelling/ner)、POS-Tagging等）、中文分词、[文本分类](reproduction/text_classification)、[Matching](reproduction/matching)、[指代消解](reproduction/coreference_resolution)、[摘要](reproduction/Summarization)等任务； 也可以使用它快速构建许多复杂的网络模型，进行科研。它具有如下的特性： -- 统一的Tabular式数据容器，让数据预处理过程简洁明了。内置多种数据集的DataSet Loader，省去预处理代码; +- 统一的Tabular式数据容器，让数据预处理过程简洁明了。内置多种数据集的Loader和Pipe，省去预处理代码; - 多种训练、测试组件，例如训练器Trainer；测试器Tester；以及各种评测metrics等等; - 各种方便的NLP工具，例如预处理embedding加载（包括ELMo和BERT）; 中间数据cache等; +- 部分[数据集与预训练模型](https://docs.qq.com/sheet/DVnpkTnF6VW9UeXdh?c=A1A0A0)的自动下载 - 详尽的中文[文档](https://fastnlp.readthedocs.io/)、[教程](https://fastnlp.readthedocs.io/zh/latest/user/tutorials.html)以供查阅; - 提供诸多高级模块，例如Variational LSTM, Transformer, CRF等; - 在序列标注、中文分词、文本分类、Matching、指代消解、摘要等任务上封装了各种模型可供直接使用，详细内容见 [reproduction](reproduction) 部分; @@ -36,7 +37,7 @@ pip install fastNLP python -m spacy download en ``` -目前使用pip安装fastNLP的版本是0.4.1，有较多功能仍未更新，最新内容以master分支为准。 +目前使用pypi安装fastNLP的版本是0.4.1，有较多功能仍未更新，最新内容以master分支为准。 fastNLP0.5.0版本将在近期推出，请密切关注。 @@ -44,7 +45,7 @@ fastNLP0.5.0版本将在近期推出，请密切关注。 - [0. 快速入门](https://fastnlp.readthedocs.io/zh/latest/user/quickstart.html) - [1. 使用DataSet预处理文本](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_1_data_preprocess.html) -- [2. 使用DataSetLoader加载数据集](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_2_load_dataset.html) +- [2. 使用Loader和Pipe加载并处理数据集](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_2_load_dataset.html) - [3. 使用Embedding模块将文本转成向量](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_3_embedding.html) - [4. 动手实现一个文本分类器I-使用Trainer和Tester快速训练和测试](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_4_loss_optimizer.html) - [5. 动手实现一个文本分类器II-使用DataSetIter实现自定义训练过程](https://fastnlp.readthedocs.io/zh/latest/tutorials/tutorial_5_datasetiter.html) @@ -118,7 +119,7 @@ fastNLP的大致工作流程如上图所示，而项目结构如下： fastNLP.io - 实现了读写功能，包括数据读入，模型读写等 + 实现了读写功能，包括数据读入与预处理，模型读写，自动下载等 diff --git a/docs/Makefile b/docs/Makefile index 2b4de2d8..b41beb44 100644 --- a/docs/Makefile +++ b/docs/Makefile @@ -14,13 +14,13 @@ help: @$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O) apidoc: - $(SPHINXAPIDOC) -efM -o source ../$(SPHINXPROJ) + $(SPHINXAPIDOC) -efM -o source ../$(SPHINXPROJ) && python3 format.py server: cd build/html && python -m http.server dev: - rm -rf build/html && make html && make server + rm -rf build && make html && make server .PHONY: help Makefile diff --git a/docs/count.py b/docs/count.py new file mode 100644 index 00000000..7118216a --- /dev/null +++ b/docs/count.py @@ -0,0 +1,142 @@ +import inspect +import os +import sys + + +def _colored_string(string: str, color: str or int) -> str: + """在终端中显示一串有颜色的文字 + :param string: 在终端中显示的文字 + :param color: 文字的颜色 + :return: + """ + if isinstance(color, str): + color = { + "black": 30, "Black": 30, "BLACK": 30, + "red": 31, "Red": 31, "RED": 31, + "green": 32, "Green": 32, "GREEN": 32, + "yellow": 33, "Yellow": 33, "YELLOW": 33, + "blue": 34, "Blue": 34, "BLUE": 34, + "purple": 35, "Purple": 35, "PURPLE": 35, + "cyan": 36, "Cyan": 36, "CYAN": 36, + "white": 37, "White": 37, "WHITE": 37 + }[color] + return "\033[%dm%s\033[0m" % (color, string) + + +def gr(string, flag): + if flag: + return _colored_string(string, "green") + else: + return _colored_string(string, "red") + + +def find_all_modules(): + modules = {} + children = {} + to_doc = set() + root = '../fastNLP' + for path, dirs, files in os.walk(root): + for file in files: + if file.endswith('.py'): + name = ".".join(path.split('/')[1:]) + if file.split('.')[0] != "__init__": + name = name + '.' + file.split('.')[0] + __import__(name) + m = sys.modules[name] + modules[name] = m + try: + m.__all__ + except: + print(name, "__all__ missing") + continue + if m.__doc__ is None: + print(name, "__doc__ missing") + continue + if "undocumented" not in m.__doc__: + to_doc.add(name) + for module in to_doc: + t = ".".join(module.split('.')[:-1]) + if t in to_doc: + if t not in children: + children[t] = set() + children[t].add(module) + for m in children: + children[m] = sorted(children[m]) + return modules, to_doc, children + + +def create_rst_file(modules, name, children): + m = modules[name] + with open("./source/" + name + ".rst", "w") as fout: + t = "=" * len(name) + fout.write(name + "\n") + fout.write(t + "\n") + fout.write("\n") + fout.write(".. automodule:: " + name + "\n") + if name != "fastNLP.core" and len(m.__all__) > 0: + fout.write(" :members: " + ", ".join(m.__all__) + "\n") + short = name[len("fastNLP."):] + if not (short.startswith('models') or short.startswith('modules') or short.startswith('embeddings')): + fout.write(" :inherited-members:\n") + fout.write("\n") + if name in children: + fout.write("子模块\n------\n\n.. toctree::\n :maxdepth: 1\n\n") + for module in children[name]: + fout.write(" " + module + "\n") + + +def check_file(m, name): + names = name.split('.') + test_name = "test." + ".".join(names[1:-1]) + ".test_" + names[-1] + try: + __import__(test_name) + tm = sys.modules[test_name] + except ModuleNotFoundError: + tm = None + tested = tm is not None + funcs = {} + classes = {} + for item, obj in inspect.getmembers(m): + if inspect.isclass(obj) and obj.__module__ == name and not obj.__name__.startswith('_'): + this = (obj.__doc__ is not None, tested and obj.__name__ in dir(tm), {}) + for i in dir(obj): + func = getattr(obj, i) + if inspect.isfunction(func) and not i.startswith('_'): + this[2][i] = (func.__doc__ is not None, False) + classes[obj.__name__] = this + if inspect.isfunction(obj) and obj.__module__ == name and not obj.__name__.startswith('_'): + this = (obj.__doc__ is not None, tested and obj.__name__ in dir(tm)) # docs + funcs[obj.__name__] = this + return funcs, classes + + +def check_files(modules, out=sys.stdout): + for name in sorted(modules.keys()): + print(name, file=out) + funcs, classes = check_file(modules[name], name) + for f in funcs: + print("%-30s \t %s \t %s" % (f, gr("文档", funcs[f][0]), gr("测试", funcs[f][1])), file=out) + for c in classes: + print("%-30s \t %s \t %s" % (c, gr("文档", classes[c][0]), gr("测试", classes[c][1])), file=out) + methods = classes[c][2] + for f in methods: + print(" %-28s \t %s" % (f, gr("文档", methods[f][0])), file=out) + print(file=out) + + +def main(): + sys.path.append("..") + print(_colored_string('Getting modules...', "Blue")) + modules, to_doc, children = find_all_modules() + print(_colored_string('Done!', "Green")) + print(_colored_string('Creating rst files...', "Blue")) + for name in to_doc: + create_rst_file(modules, name, children) + print(_colored_string('Done!', "Green")) + print(_colored_string('Checking all files...', "Blue")) + check_files(modules) + print(_colored_string('Done!', "Green")) + + +if __name__ == "__main__": + main() diff --git a/docs/source/conf.py b/docs/source/conf.py index 2e10bc89..7536ee32 100644 --- a/docs/source/conf.py +++ b/docs/source/conf.py @@ -48,12 +48,14 @@ extensions = [ autodoc_default_options = { 'member-order': 'bysource', 'special-members': '__init__', - 'undoc-members': True, + 'undoc-members': False, } +autoclass_content = "class" + # Add any paths that contain templates here, relative to this directory. templates_path = ['_templates'] - +# template_bridge # The suffix(es) of source filenames. # You can specify multiple suffix as a list of string: # @@ -113,7 +115,7 @@ html_static_path = ['_static'] # -- Options for HTMLHelp output --------------------------------------------- # Output file base name for HTML help builder. -htmlhelp_basename = 'fastNLPdoc' +htmlhelp_basename = 'fastNLP doc' # -- Options for LaTeX output ------------------------------------------------ @@ -166,10 +168,12 @@ texinfo_documents = [ # -- Extension configuration ------------------------------------------------- def maybe_skip_member(app, what, name, obj, skip, options): - if name.startswith("_"): - return True if obj.__doc__ is None: return True + if name == "__init__": + return False + if name.startswith("_"): + return True return False diff --git a/docs/source/fastNLP.core.batch.rst b/docs/source/fastNLP.core.batch.rst index 03008b52..50ad6fed 100644 --- a/docs/source/fastNLP.core.batch.rst +++ b/docs/source/fastNLP.core.batch.rst @@ -2,6 +2,6 @@ fastNLP.core.batch ================== .. automodule:: fastNLP.core.batch - :members: - :undoc-members: - :show-inheritance: + :members: BatchIter, DataSetIter, TorchLoaderIter + :inherited-members: + diff --git a/docs/source/fastNLP.core.callback.rst b/docs/source/fastNLP.core.callback.rst index 74a7825d..d37ddb11 100644 --- a/docs/source/fastNLP.core.callback.rst +++ b/docs/source/fastNLP.core.callback.rst @@ -2,6 +2,6 @@ fastNLP.core.callback ===================== .. automodule:: fastNLP.core.callback - :members: - :undoc-members: - :show-inheritance: + :members: Callback, GradientClipCallback, EarlyStopCallback, FitlogCallback, EvaluateCallback, LRScheduler, ControlC, LRFinder, TensorboardCallback, WarmupCallback, SaveModelCallback, EchoCallback, TesterCallback, CallbackException, EarlyStopError + :inherited-members: + diff --git a/docs/source/fastNLP.core.const.rst b/docs/source/fastNLP.core.const.rst index 330a8883..82a1992e 100644 --- a/docs/source/fastNLP.core.const.rst +++ b/docs/source/fastNLP.core.const.rst @@ -2,6 +2,6 @@ fastNLP.core.const ================== .. automodule:: fastNLP.core.const - :members: - :undoc-members: - :show-inheritance: + :members: Const + :inherited-members: + diff --git a/docs/source/fastNLP.core.dataset.rst b/docs/source/fastNLP.core.dataset.rst index 1ad94bb6..e13d7f1c 100644 --- a/docs/source/fastNLP.core.dataset.rst +++ b/docs/source/fastNLP.core.dataset.rst @@ -2,6 +2,6 @@ fastNLP.core.dataset ==================== .. automodule:: fastNLP.core.dataset - :members: - :undoc-members: - :show-inheritance: + :members: DataSet + :inherited-members: + diff --git a/docs/source/fastNLP.core.field.rst b/docs/source/fastNLP.core.field.rst index 7fc099c9..73dad8af 100644 --- a/docs/source/fastNLP.core.field.rst +++ b/docs/source/fastNLP.core.field.rst @@ -2,6 +2,6 @@ fastNLP.core.field ================== .. automodule:: fastNLP.core.field - :members: - :undoc-members: - :show-inheritance: + :members: Padder, AutoPadder, EngChar2DPadder + :inherited-members: + diff --git a/docs/source/fastNLP.core.instance.rst b/docs/source/fastNLP.core.instance.rst index 6e496ac1..010567b9 100644 --- a/docs/source/fastNLP.core.instance.rst +++ b/docs/source/fastNLP.core.instance.rst @@ -2,6 +2,6 @@ fastNLP.core.instance ===================== .. automodule:: fastNLP.core.instance - :members: - :undoc-members: - :show-inheritance: + :members: Instance + :inherited-members: + diff --git a/docs/source/fastNLP.core.losses.rst b/docs/source/fastNLP.core.losses.rst index 8e63dfa1..daf246f8 100644 --- a/docs/source/fastNLP.core.losses.rst +++ b/docs/source/fastNLP.core.losses.rst @@ -2,6 +2,6 @@ fastNLP.core.losses =================== .. automodule:: fastNLP.core.losses - :members: - :undoc-members: - :show-inheritance: + :members: LossBase, LossFunc, LossInForward, CrossEntropyLoss, BCELoss, L1Loss, NLLLoss + :inherited-members: + diff --git a/docs/source/fastNLP.core.metrics.rst b/docs/source/fastNLP.core.metrics.rst index d3b87bb8..96748a78 100644 --- a/docs/source/fastNLP.core.metrics.rst +++ b/docs/source/fastNLP.core.metrics.rst @@ -2,6 +2,6 @@ fastNLP.core.metrics ==================== .. automodule:: fastNLP.core.metrics - :members: - :undoc-members: - :show-inheritance: + :members: MetricBase, AccuracyMetric, SpanFPreRecMetric, ExtractiveQAMetric + :inherited-members: + diff --git a/docs/source/fastNLP.core.optimizer.rst b/docs/source/fastNLP.core.optimizer.rst index c80be53f..44e45c4f 100644 --- a/docs/source/fastNLP.core.optimizer.rst +++ b/docs/source/fastNLP.core.optimizer.rst @@ -2,6 +2,6 @@ fastNLP.core.optimizer ====================== .. automodule:: fastNLP.core.optimizer - :members: - :undoc-members: - :show-inheritance: + :members: Optimizer, SGD, Adam, AdamW + :inherited-members: + diff --git a/docs/source/fastNLP.core.rst b/docs/source/fastNLP.core.rst index cacc6622..15fe29d5 100644 --- a/docs/source/fastNLP.core.rst +++ b/docs/source/fastNLP.core.rst @@ -2,12 +2,9 @@ fastNLP.core ============ .. automodule:: fastNLP.core - :members: - :undoc-members: - :show-inheritance: 子模块 ----------- +------ .. toctree:: :maxdepth: 1 diff --git a/docs/source/fastNLP.core.sampler.rst b/docs/source/fastNLP.core.sampler.rst index 0110f0c0..56291894 100644 --- a/docs/source/fastNLP.core.sampler.rst +++ b/docs/source/fastNLP.core.sampler.rst @@ -2,6 +2,6 @@ fastNLP.core.sampler ==================== .. automodule:: fastNLP.core.sampler - :members: - :undoc-members: - :show-inheritance: + :members: Sampler, BucketSampler, SequentialSampler, RandomSampler + :inherited-members: + diff --git a/docs/source/fastNLP.core.tester.rst b/docs/source/fastNLP.core.tester.rst index 4d71a27b..90ec2a88 100644 --- a/docs/source/fastNLP.core.tester.rst +++ b/docs/source/fastNLP.core.tester.rst @@ -2,6 +2,6 @@ fastNLP.core.tester =================== .. automodule:: fastNLP.core.tester - :members: - :undoc-members: - :show-inheritance: + :members: Tester + :inherited-members: + diff --git a/docs/source/fastNLP.core.trainer.rst b/docs/source/fastNLP.core.trainer.rst index 60bf2d5b..92c08718 100644 --- a/docs/source/fastNLP.core.trainer.rst +++ b/docs/source/fastNLP.core.trainer.rst @@ -2,6 +2,6 @@ fastNLP.core.trainer ==================== .. automodule:: fastNLP.core.trainer - :members: - :undoc-members: - :show-inheritance: + :members: Trainer + :inherited-members: + diff --git a/docs/source/fastNLP.core.utils.rst b/docs/source/fastNLP.core.utils.rst index 3f80b4e8..027a43e9 100644 --- a/docs/source/fastNLP.core.utils.rst +++ b/docs/source/fastNLP.core.utils.rst @@ -2,6 +2,6 @@ fastNLP.core.utils ================== .. automodule:: fastNLP.core.utils - :members: - :undoc-members: - :show-inheritance: + :members: cache_results, seq_len_to_mask, get_seq_len + :inherited-members: + diff --git a/docs/source/fastNLP.core.vocabulary.rst b/docs/source/fastNLP.core.vocabulary.rst index ba9598b9..ac07a8c6 100644 --- a/docs/source/fastNLP.core.vocabulary.rst +++ b/docs/source/fastNLP.core.vocabulary.rst @@ -2,6 +2,6 @@ fastNLP.core.vocabulary ======================= .. automodule:: fastNLP.core.vocabulary - :members: - :undoc-members: - :show-inheritance: + :members: Vocabulary, VocabularyOption + :inherited-members: + diff --git a/docs/source/fastNLP.embeddings.bert_embedding.rst b/docs/source/fastNLP.embeddings.bert_embedding.rst index 24ceff1c..1b59dc35 100644 --- a/docs/source/fastNLP.embeddings.bert_embedding.rst +++ b/docs/source/fastNLP.embeddings.bert_embedding.rst @@ -1,7 +1,6 @@ -fastNLP.embeddings.bert\_embedding -================================== +fastNLP.embeddings.bert_embedding +================================= .. automodule:: fastNLP.embeddings.bert_embedding - :members: - :undoc-members: - :show-inheritance: + :members: BertEmbedding, BertWordPieceEncoder + diff --git a/docs/source/fastNLP.embeddings.char_embedding.rst b/docs/source/fastNLP.embeddings.char_embedding.rst index 501089d8..bc8d64f9 100644 --- a/docs/source/fastNLP.embeddings.char_embedding.rst +++ b/docs/source/fastNLP.embeddings.char_embedding.rst @@ -1,7 +1,6 @@ -fastNLP.embeddings.char\_embedding -================================== +fastNLP.embeddings.char_embedding +================================= .. automodule:: fastNLP.embeddings.char_embedding - :members: - :undoc-members: - :show-inheritance: + :members: CNNCharEmbedding, LSTMCharEmbedding + diff --git a/docs/source/fastNLP.embeddings.contextual_embedding.rst b/docs/source/fastNLP.embeddings.contextual_embedding.rst new file mode 100644 index 00000000..74e5f5be --- /dev/null +++ b/docs/source/fastNLP.embeddings.contextual_embedding.rst @@ -0,0 +1,6 @@ +fastNLP.embeddings.contextual_embedding +======================================= + +.. automodule:: fastNLP.embeddings.contextual_embedding + :members: ContextualEmbedding + diff --git a/docs/source/fastNLP.embeddings.elmo_embedding.rst b/docs/source/fastNLP.embeddings.elmo_embedding.rst index 76669ee3..b8c6d41c 100644 --- a/docs/source/fastNLP.embeddings.elmo_embedding.rst +++ b/docs/source/fastNLP.embeddings.elmo_embedding.rst @@ -1,7 +1,6 @@ -fastNLP.embeddings.elmo\_embedding -================================== +fastNLP.embeddings.elmo_embedding +================================= .. automodule:: fastNLP.embeddings.elmo_embedding - :members: - :undoc-members: - :show-inheritance: + :members: ElmoEmbedding + diff --git a/docs/source/fastNLP.embeddings.embedding.rst b/docs/source/fastNLP.embeddings.embedding.rst index 5960d2cd..6793446b 100644 --- a/docs/source/fastNLP.embeddings.embedding.rst +++ b/docs/source/fastNLP.embeddings.embedding.rst @@ -2,6 +2,5 @@ fastNLP.embeddings.embedding ============================ .. automodule:: fastNLP.embeddings.embedding - :members: - :undoc-members: - :show-inheritance: + :members: Embedding, TokenEmbedding + diff --git a/docs/source/fastNLP.embeddings.rst b/docs/source/fastNLP.embeddings.rst index 6b168906..f4f4a3e0 100644 --- a/docs/source/fastNLP.embeddings.rst +++ b/docs/source/fastNLP.embeddings.rst @@ -2,18 +2,17 @@ fastNLP.embeddings ================== .. automodule:: fastNLP.embeddings - :members: - :undoc-members: - :show-inheritance: + :members: Embedding, TokenEmbedding, StaticEmbedding, ElmoEmbedding, BertEmbedding, BertWordPieceEncoder, StackEmbedding, LSTMCharEmbedding, CNNCharEmbedding, get_embeddings 子模块 ----------- +------ .. toctree:: :maxdepth: 1 fastNLP.embeddings.bert_embedding fastNLP.embeddings.char_embedding + fastNLP.embeddings.contextual_embedding fastNLP.embeddings.elmo_embedding fastNLP.embeddings.embedding fastNLP.embeddings.stack_embedding diff --git a/docs/source/fastNLP.embeddings.stack_embedding.rst b/docs/source/fastNLP.embeddings.stack_embedding.rst index 4d2115f7..a07d1ef5 100644 --- a/docs/source/fastNLP.embeddings.stack_embedding.rst +++ b/docs/source/fastNLP.embeddings.stack_embedding.rst @@ -1,7 +1,6 @@ -fastNLP.embeddings.stack\_embedding -=================================== +fastNLP.embeddings.stack_embedding +================================== .. automodule:: fastNLP.embeddings.stack_embedding - :members: - :undoc-members: - :show-inheritance: + :members: StackEmbedding + diff --git a/docs/source/fastNLP.embeddings.static_embedding.rst b/docs/source/fastNLP.embeddings.static_embedding.rst index e46de81a..219ce0e5 100644 --- a/docs/source/fastNLP.embeddings.static_embedding.rst +++ b/docs/source/fastNLP.embeddings.static_embedding.rst @@ -1,7 +1,6 @@ -fastNLP.embeddings.static\_embedding -==================================== +fastNLP.embeddings.static_embedding +=================================== .. automodule:: fastNLP.embeddings.static_embedding - :members: - :undoc-members: - :show-inheritance: + :members: StaticEmbedding + diff --git a/docs/source/fastNLP.embeddings.utils.rst b/docs/source/fastNLP.embeddings.utils.rst index 263bfbd6..077487c1 100644 --- a/docs/source/fastNLP.embeddings.utils.rst +++ b/docs/source/fastNLP.embeddings.utils.rst @@ -2,6 +2,5 @@ fastNLP.embeddings.utils ======================== .. automodule:: fastNLP.embeddings.utils - :members: - :undoc-members: - :show-inheritance: + :members: get_embeddings + diff --git a/docs/source/fastNLP.io.base_loader.rst b/docs/source/fastNLP.io.base_loader.rst deleted file mode 100644 index 057867f4..00000000 --- a/docs/source/fastNLP.io.base_loader.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.io.base\_loader -======================= - -.. automodule:: fastNLP.io.base_loader - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.io.data_bundle.rst b/docs/source/fastNLP.io.data_bundle.rst new file mode 100644 index 00000000..71a921f1 --- /dev/null +++ b/docs/source/fastNLP.io.data_bundle.rst @@ -0,0 +1,7 @@ +fastNLP.io.data_bundle +====================== + +.. automodule:: fastNLP.io.data_bundle + :members: DataBundle + :inherited-members: + diff --git a/docs/source/fastNLP.io.data_loader.rst b/docs/source/fastNLP.io.data_loader.rst deleted file mode 100644 index 8f990102..00000000 --- a/docs/source/fastNLP.io.data_loader.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.io.data\_loader -========================== - -.. automodule:: fastNLP.io.data_loader - :members: - :undoc-members: - :show-inheritance: \ No newline at end of file diff --git a/docs/source/fastNLP.io.dataset_loader.rst b/docs/source/fastNLP.io.dataset_loader.rst deleted file mode 100644 index e7990714..00000000 --- a/docs/source/fastNLP.io.dataset_loader.rst +++ /dev/null @@ -1,7 +0,0 @@ -fastNLP.io.dataset\_loader -========================== - -.. automodule:: fastNLP.io.dataset_loader - :members: - :undoc-members: - :show-inheritance: diff --git a/docs/source/fastNLP.io.embed_loader.rst b/docs/source/fastNLP.io.embed_loader.rst index 69e1f7ff..581f5c1b 100644 --- a/docs/source/fastNLP.io.embed_loader.rst +++ b/docs/source/fastNLP.io.embed_loader.rst @@ -1,7 +1,7 @@ -fastNLP.io.embed\_loader -======================== +fastNLP.io.embed_loader +======================= .. automodule:: fastNLP.io.embed_loader - :members: - :undoc-members: - :show-inheritance: + :members: EmbedLoader, EmbeddingOption + :inherited-members: + diff --git a/docs/source/fastNLP.io.file_utils.rst b/docs/source/fastNLP.io.file_utils.rst new file mode 100644 index 00000000..0815e068 --- /dev/null +++ b/docs/source/fastNLP.io.file_utils.rst @@ -0,0 +1,7 @@ +fastNLP.io.file_utils +===================== + +.. automodule:: fastNLP.io.file_utils + :members: cached_path, get_filepath, get_cache_path, split_filename_suffix, get_from_cache + :inherited-members: + diff --git a/docs/source/fastNLP.io.loader.rst b/docs/source/fastNLP.io.loader.rst new file mode 100644 index 00000000..c1af6c0c --- /dev/null +++ b/docs/source/fastNLP.io.loader.rst @@ -0,0 +1,7 @@ +fastNLP.io.loader +================= + +.. automodule:: fastNLP.io.loader + :members: Loader, YelpLoader, YelpFullLoader, YelpPolarityLoader, IMDBLoader, SSTLoader, SST2Loader, ChnSentiCorpLoader, ConllLoader, Conll2003Loader, Conll2003NERLoader, OntoNotesNERLoader, CTBLoader, MsraNERLoader, PeopleDailyNERLoader, WeiboNERLoader, CSVLoader, JsonLoader, CWSLoader, MNLILoader, QuoraLoader, SNLILoader, QNLILoader, RTELoader + :inherited-members: + diff --git a/docs/source/fastNLP.io.model_io.rst b/docs/source/fastNLP.io.model_io.rst index 537ce752..183122b1 100644 --- a/docs/source/fastNLP.io.model_io.rst +++ b/docs/source/fastNLP.io.model_io.rst @@ -1,7 +1,7 @@ -fastNLP.io.model\_io -==================== +fastNLP.io.model_io +=================== .. automodule:: fastNLP.io.model_io - :members: - :undoc-members: - :show-inheritance: + :members: ModelLoader, ModelSaver + :inherited-members: + diff --git a/docs/source/fastNLP.io.pipe.rst b/docs/source/fastNLP.io.pipe.rst new file mode 100644 index 00000000..3ef9b5a8 --- /dev/null +++ b/docs/source/fastNLP.io.pipe.rst @@ -0,0 +1,7 @@ +fastNLP.io.pipe +=============== + +.. automodule:: fastNLP.io.pipe + :members: Pipe, CWSPipe, YelpFullPipe, YelpPolarityPipe, SSTPipe, SST2Pipe, IMDBPipe, ChnSentiCorpPipe, Conll2003NERPipe, OntoNotesNERPipe, MsraNERPipe, WeiboNERPipe, PeopleDailyPipe, Conll2003Pipe, MatchingBertPipe, RTEBertPipe, SNLIBertPipe, QuoraBertPipe, QNLIBertPipe, MNLIBertPipe, MatchingPipe, RTEPipe, SNLIPipe, QuoraPipe, QNLIPipe, MNLIPipe + :inherited-members: + diff --git a/docs/source/fastNLP.io.rst b/docs/source/fastNLP.io.rst index a97ed67d..7118039d 100644 --- a/docs/source/fastNLP.io.rst +++ b/docs/source/fastNLP.io.rst @@ -2,18 +2,19 @@ fastNLP.io ========== .. automodule:: fastNLP.io - :members: - :undoc-members: - :show-inheritance: + :members: DataBundle, EmbedLoader, Loader, YelpLoader, YelpFullLoader, YelpPolarityLoader, IMDBLoader, SSTLoader, SST2Loader, ChnSentiCorpLoader, ConllLoader, Conll2003Loader, Conll2003NERLoader, OntoNotesNERLoader, CTBLoader, MsraNERLoader, WeiboNERLoader, PeopleDailyNERLoader, CSVLoader, JsonLoader, CWSLoader, MNLILoader, QuoraLoader, SNLILoader, QNLILoader, RTELoader, Pipe, YelpFullPipe, YelpPolarityPipe, SSTPipe, SST2Pipe, IMDBPipe, ChnSentiCorpPipe, Conll2003Pipe, Conll2003NERPipe, OntoNotesNERPipe, MsraNERPipe, PeopleDailyPipe, WeiboNERPipe, CWSPipe, MatchingBertPipe, RTEBertPipe, SNLIBertPipe, QuoraBertPipe, QNLIBertPipe, MNLIBertPipe, MatchingPipe, RTEPipe, SNLIPipe, QuoraPipe, QNLIPipe, MNLIPipe, ModelLoader, ModelSaver + :inherited-members: 子模块 ----------- +------ .. toctree:: :maxdepth: 1 - fastNLP.io.base_loader + fastNLP.io.data_bundle fastNLP.io.embed_loader - fastNLP.io.dataset_loader - fastNLP.io.data_loader + fastNLP.io.file_utils + fastNLP.io.loader fastNLP.io.model_io + fastNLP.io.pipe + fastNLP.io.utils diff --git a/docs/source/fastNLP.io.utils.rst b/docs/source/fastNLP.io.utils.rst new file mode 100644 index 00000000..3bff3c45 --- /dev/null +++ b/docs/source/fastNLP.io.utils.rst @@ -0,0 +1,7 @@ +fastNLP.io.utils +================ + +.. automodule:: fastNLP.io.utils + :members: check_loader_paths + :inherited-members: + diff --git a/docs/source/fastNLP.models.bert.rst b/docs/source/fastNLP.models.bert.rst new file mode 100644 index 00000000..b0c813f9 --- /dev/null +++ b/docs/source/fastNLP.models.bert.rst @@ -0,0 +1,6 @@ +fastNLP.models.bert +=================== + +.. automodule:: fastNLP.models.bert + :members: BertForSequenceClassification, BertForSentenceMatching, BertForMultipleChoice, BertForTokenClassification, BertForQuestionAnswering + diff --git a/docs/source/fastNLP.models.biaffine_parser.rst b/docs/source/fastNLP.models.biaffine_parser.rst index f19504e8..395638fe 100644 --- a/docs/source/fastNLP.models.biaffine_parser.rst +++ b/docs/source/fastNLP.models.biaffine_parser.rst @@ -1,7 +1,6 @@ -fastNLP.models.biaffine\_parser -=============================== +fastNLP.models.biaffine_parser +============================== .. automodule:: fastNLP.models.biaffine_parser - :members: - :undoc-members: - :show-inheritance: + :members: BiaffineParser, GraphParser + diff --git a/docs/source/fastNLP.models.cnn_text_classification.rst b/docs/source/fastNLP.models.cnn_text_classification.rst index eacf6916..e9ed7ee1 100644 --- a/docs/source/fastNLP.models.cnn_text_classification.rst +++ b/docs/source/fastNLP.models.cnn_text_classification.rst @@ -1,7 +1,6 @@ -fastNLP.models.cnn\_text\_classification -======================================== +fastNLP.models.cnn_text_classification +====================================== .. automodule:: fastNLP.models.cnn_text_classification - :members: - :undoc-members: - :show-inheritance: + :members: CNNText + diff --git a/docs/source/fastNLP.models.rst b/docs/source/fastNLP.models.rst index 2ea546e2..21cf41a7 100644 --- a/docs/source/fastNLP.models.rst +++ b/docs/source/fastNLP.models.rst @@ -2,16 +2,15 @@ fastNLP.models ============== .. automodule:: fastNLP.models - :members: - :undoc-members: - :show-inheritance: + :members: CNNText, SeqLabeling, AdvSeqLabel, ESIM, StarTransEnc, STSeqLabel, STNLICls, STSeqCls, BiaffineParser, GraphParser, BertForSequenceClassification, BertForSentenceMatching, BertForMultipleChoice, BertForTokenClassification, BertForQuestionAnswering 子模块 ----------- +------ .. toctree:: :maxdepth: 1 + fastNLP.models.bert fastNLP.models.biaffine_parser fastNLP.models.cnn_text_classification fastNLP.models.sequence_labeling diff --git a/docs/source/fastNLP.models.sequence_labeling.rst b/docs/source/fastNLP.models.sequence_labeling.rst index 85e28f06..dcd1300e 100644 --- a/docs/source/fastNLP.models.sequence_labeling.rst +++ b/docs/source/fastNLP.models.sequence_labeling.rst @@ -1,7 +1,6 @@ -fastNLP.models.sequence\_labeling -================================= +fastNLP.models.sequence_labeling +================================ .. automodule:: fastNLP.models.sequence_labeling - :members: - :undoc-members: - :show-inheritance: + :members: SeqLabeling, AdvSeqLabel, BiLSTMCRF + diff --git a/docs/source/fastNLP.models.snli.rst b/docs/source/fastNLP.models.snli.rst index 3b9b555c..eed02139 100644 --- a/docs/source/fastNLP.models.snli.rst +++ b/docs/source/fastNLP.models.snli.rst @@ -2,6 +2,5 @@ fastNLP.models.snli =================== .. automodule:: fastNLP.models.snli - :members: - :undoc-members: - :show-inheritance: + :members: ESIM + diff --git a/docs/source/fastNLP.models.star_transformer.rst b/docs/source/fastNLP.models.star_transformer.rst index 69d5c5b2..80ab5b33 100644 --- a/docs/source/fastNLP.models.star_transformer.rst +++ b/docs/source/fastNLP.models.star_transformer.rst @@ -1,7 +1,6 @@ -fastNLP.models.star\_transformer -================================ +fastNLP.models.star_transformer +=============================== .. automodule:: fastNLP.models.star_transformer - :members: - :undoc-members: - :show-inheritance: + :members: StarTransEnc, STNLICls, STSeqCls, STSeqLabel + diff --git a/docs/source/fastNLP.modules.decoder.rst b/docs/source/fastNLP.modules.decoder.rst index ecc2adbd..de6e0d9d 100644 --- a/docs/source/fastNLP.modules.decoder.rst +++ b/docs/source/fastNLP.modules.decoder.rst @@ -2,7 +2,5 @@ fastNLP.modules.decoder ======================= .. automodule:: fastNLP.modules.decoder - :members: - :undoc-members: - :show-inheritance: + :members: MLP, ConditionalRandomField, viterbi_decode, allowed_transitions diff --git a/docs/source/fastNLP.modules.encoder.rst b/docs/source/fastNLP.modules.encoder.rst index 0562f12d..fceabbdb 100644 --- a/docs/source/fastNLP.modules.encoder.rst +++ b/docs/source/fastNLP.modules.encoder.rst @@ -2,6 +2,5 @@ fastNLP.modules.encoder ======================= .. automodule:: fastNLP.modules.encoder - :members: - :undoc-members: - :show-inheritance: + :members: ConvolutionCharEncoder, LSTMCharEncoder, ConvMaxpool, LSTM, StarTransformer, TransformerEncoder, VarRNN, VarLSTM, VarGRU, MaxPool, MaxPoolWithMask, AvgPool, AvgPoolWithMask, MultiHeadAttention + diff --git a/docs/source/fastNLP.modules.rst b/docs/source/fastNLP.modules.rst index 646ef2d3..b7c259ed 100644 --- a/docs/source/fastNLP.modules.rst +++ b/docs/source/fastNLP.modules.rst @@ -2,16 +2,14 @@ fastNLP.modules =============== .. automodule:: fastNLP.modules - :members: - :undoc-members: - :show-inheritance: + :members: ConvolutionCharEncoder, LSTMCharEncoder, ConvMaxpool, LSTM, StarTransformer, TransformerEncoder, VarRNN, VarLSTM, VarGRU, MaxPool, MaxPoolWithMask, AvgPool, AvgPoolWithMask, MultiHeadAttention, MLP, ConditionalRandomField, viterbi_decode, allowed_transitions, TimestepDropout 子模块 ------------ +------ .. toctree:: - :titlesonly: :maxdepth: 1 fastNLP.modules.decoder - fastNLP.modules.encoder \ No newline at end of file + fastNLP.modules.encoder + fastNLP.modules.utils diff --git a/docs/source/fastNLP.modules.utils.rst b/docs/source/fastNLP.modules.utils.rst new file mode 100644 index 00000000..101a0f45 --- /dev/null +++ b/docs/source/fastNLP.modules.utils.rst @@ -0,0 +1,6 @@ +fastNLP.modules.utils +===================== + +.. automodule:: fastNLP.modules.utils + :members: initial_parameter, summary + diff --git a/docs/source/fastNLP.rst b/docs/source/fastNLP.rst index 0057a184..e01817f7 100644 --- a/docs/source/fastNLP.rst +++ b/docs/source/fastNLP.rst @@ -1,13 +1,12 @@ -API 文档 -=============== +fastNLP +======= .. automodule:: fastNLP - :members: - :undoc-members: - :show-inheritance: + :members: Instance, FieldArray, DataSetIter, BatchIter, TorchLoaderIter, Vocabulary, DataSet, Const, Trainer, Tester, Callback, GradientClipCallback, EarlyStopCallback, TensorboardCallback, LRScheduler, ControlC, LRFinder, Padder, AutoPadder, EngChar2DPadder, AccuracyMetric, SpanFPreRecMetric, ExtractiveQAMetric, Optimizer, SGD, Adam, AdamW, Sampler, SequentialSampler, BucketSampler, RandomSampler, LossFunc, CrossEntropyLoss, L1Loss, BCELoss, NLLLoss, LossInForward, cache_results, logger + :inherited-members: -内部模块 ------------ +子模块 +------ .. toctree:: :maxdepth: 1 diff --git a/docs/source/modules.rst b/docs/source/modules.rst index 9ca3c7f3..e9a92cb7 100644 --- a/docs/source/modules.rst +++ b/docs/source/modules.rst @@ -2,7 +2,6 @@ fastNLP ======= .. toctree:: - :titlesonly: :maxdepth: 4 fastNLP diff --git a/docs/source/tutorials/tutorial_2_load_dataset.rst b/docs/source/tutorials/tutorial_2_load_dataset.rst index 4fa4a84d..17ad6baf 100644 --- a/docs/source/tutorials/tutorial_2_load_dataset.rst +++ b/docs/source/tutorials/tutorial_2_load_dataset.rst @@ -1,57 +1,53 @@ -================================= -使用DataSetLoader加载数据集 -================================= +======================================= +使用Loader和Pipe加载并处理数据集 +======================================= 这一部分是一个关于如何加载数据集的教程 教程目录： - - `Part I: 数据集容器`_ - - `Part II: 数据集的使用方式`_ - - `Part III: 不同数据类型的DataSetLoader`_ - - `Part IV: DataSetLoader举例`_ - - `Part V: fastNLP封装好的数据集加载器`_ + - `Part I: 数据集容器DataBundle`_ + - `Part II: 加载数据集的基类Loader`_ + - `Part III: 不同格式类型的基础Loader`_ + - `Part IV: 使用Pipe对数据集进行预处理`_ + - `Part V: fastNLP封装好的Loader和Pipe`_ ----------------------------- -Part I: 数据集容器 ----------------------------- +------------------------------------ +Part I: 数据集容器DataBundle +------------------------------------ -在fastNLP中，我们使用 :class:`~fastNLP.io.base_loader.DataBundle` 来存储数据集信息。 -:class:`~fastNLP.io.base_loader.DataBundle` 类包含了两个重要内容： `datasets` 和 `vocabs` 。 +在fastNLP中，我们使用 :class:`~fastNLP.io.data_bundle.DataBundle` 来存储数据集信息。 +:class:`~fastNLP.io.data_bundle.DataBundle` 类包含了两个重要内容： `datasets` 和 `vocabs` 。 `datasets` 是一个 `key` 为数据集名称（如 `train` ， `dev` ，和 `test` 等）， `value` 为 :class:`~fastNLP.DataSet` 的字典。 `vocabs` 是一个 `key` 为词表名称（如 :attr:`fastNLP.Const.INPUT` 表示输入文本的词表名称， :attr:`fastNLP.Const.TARGET` 表示目标 的真实标签词表的名称，等等）， `value` 为词表内容（ :class:`~fastNLP.Vocabulary` ）的字典。 ----------------------------- -Part II: 数据集的使用方式 ----------------------------- +------------------------------------- +Part II: 加载数据集的基类Loader +------------------------------------- -在fastNLP中，我们采用 :class:`~fastNLP.io.base_loader.DataSetLoader` 来作为加载数据集的基类。 -:class:`~fastNLP.io.base_loader.DataSetLoader` 定义了各种DataSetLoader所需的API接口，开发者应该继承它实现各种的DataSetLoader。 -在各种数据集的DataSetLoader当中，至少应该编写如下内容: +在fastNLP中，我们采用 :class:`~fastNLP.io.loader.Loader` 来作为加载数据集的基类。 +:class:`~fastNLP.io.loader.Loader` 定义了各种Loader所需的API接口，开发者应该继承它实现各种的Loader。 +在各种数据集的Loader当中，至少应该编写如下内容: - - _load 函数：从一个数据文件中读取数据到一个 :class:`~fastNLP.DataSet` - - load 函数（可以使用基类的方法）：从一个或多个数据文件中读取数据到一个或多个 :class:`~fastNLP.DataSet` - - process 函数：一个或多个从数据文件中读取数据，并处理成可以训练的 :class:`~fastNLP.io.DataBundle` + - _load 函数：从一个数据文件中读取数据，返回一个 :class:`~fastNLP.DataSet` + - load 函数：从文件或者文件夹中读取数据并组装成 :class:`~fastNLP.io.data_bundle.DataBundle` - **\*process函数中可以调用load函数或_load函数** - -DataSetLoader的_load或者load函数返回的 :class:`~fastNLP.DataSet` 当中，内容为数据集的文本信息，process函数返回的 -:class:`~fastNLP.io.DataBundle` 当中， `datasets` 的内容为已经index好的、可以直接被 :class:`~fastNLP.Trainer` -接受的内容。 +Loader的load函数返回的 :class:`~fastNLP.io.data_bundle.DataBundle` 里面包含了数据集的原始数据。 -------------------------------------------------------- -Part III: 不同数据类型的DataSetLoader +Part III: 不同格式类型的基础Loader -------------------------------------------------------- -:class:`~fastNLP.io.dataset_loader.CSVLoader` +:class:`~fastNLP.io.loader.CSVLoader` 读取CSV类型的数据集文件。例子如下： .. code-block:: python + from fastNLP.io.loader import CSVLoader data_set_loader = CSVLoader( headers=('words', 'target'), sep='\t' ) @@ -67,17 +63,18 @@ Part III: 不同数据类型的DataSetLoader The performances are an absolute joy . 4 -:class:`~fastNLP.io.dataset_loader.JsonLoader` +:class:`~fastNLP.io.loader.JsonLoader` 读取Json类型的数据集文件，数据必须按行存储，每行是一个包含各类属性的Json对象。例子如下： .. code-block:: python - data_set_loader = JsonLoader( + from fastNLP.io.loader import JsonLoader + oader = JsonLoader( fields={'sentence1': 'words1', 'sentence2': 'words2', 'gold_label': 'target'} ) # 表示将Json对象中'sentence1'、'sentence2'和'gold_label'对应的值赋给'words1'、'words2'、'target'这三个fields - data_set = data_set_loader._load('path/to/your/file') + data_set = loader._load('path/to/your/file') 数据集内容样例如下 :: @@ -86,139 +83,68 @@ Part III: 不同数据类型的DataSetLoader {"annotator_labels": ["entailment"], "captionID": "3416050480.jpg#4", "gold_label": "entailment", "pairID": "3416050480.jpg#4r1e", "sentence1": "A person on a horse jumps over a broken down airplane.", "sentence1_binary_parse": "( ( ( A person ) ( on ( a horse ) ) ) ( ( jumps ( over ( a ( broken ( down airplane ) ) ) ) ) . ) )", "sentence1_parse": "(ROOT (S (NP (NP (DT A) (NN person)) (PP (IN on) (NP (DT a) (NN horse)))) (VP (VBZ jumps) (PP (IN over) (NP (DT a) (JJ broken) (JJ down) (NN airplane)))) (. .)))", "sentence2": "A person is outdoors, on a horse.", "sentence2_binary_parse": "( ( A person ) ( ( ( ( is outdoors ) , ) ( on ( a horse ) ) ) . ) )", "sentence2_parse": "(ROOT (S (NP (DT A) (NN person)) (VP (VBZ is) (ADVP (RB outdoors)) (, ,) (PP (IN on) (NP (DT a) (NN horse)))) (. .)))"} ------------------------------------------ -Part IV: DataSetLoader举例 +Part IV: 使用Pipe对数据集进行预处理 ------------------------------------------ -以Matching任务为例子： - - :class:`~fastNLP.io.data_loader.MatchingLoader` - 我们在fastNLP当中封装了一个Matching任务数据集的数据加载类： :class:`~fastNLP.io.data_loader.MatchingLoader` . - - 在MatchingLoader类当中我们封装了一个对数据集中的文本内容进行进一步的预处理的函数： - :meth:`~fastNLP.io.data_loader.MatchingLoader.process` - 这个函数具有各种预处理option，如： - - 是否将文本转成全小写 - - 是否需要序列长度信息，需要什么类型的序列长度信息 - - 是否需要用BertTokenizer来获取序列的WordPiece信息 - - 等等 +在fastNLP中，我们采用 :class:`~fastNLP.io.pipe.Pipe` 来作为加载数据集的基类。 +:class:`~fastNLP.io.pipe.Pipe` 定义了各种Pipe所需的API接口，开发者应该继承它实现各种的Pipe。 +在各种数据集的Pipe当中，至少应该编写如下内容: - 具体内容参见 :meth:`fastNLP.io.MatchingLoader.process` 。 + - process 函数：对输入的 :class:`~fastNLP.io.data_bundle.DataBundle` 进行处理（如构建词表、 + 将dataset的文本内容转成index等等），然后返回该 :class:`~fastNLP.io.data_bundle.DataBundle` + - process_from_file 函数：输入数据集所在文件夹，读取内容并组装成 :class:`~fastNLP.io.data_bundle.DataBundle` ， + 然后调用相对应的process函数对数据进行预处理 - :class:`~fastNLP.io.data_loader.SNLILoader` - 一个关于SNLI数据集的DataSetLoader。SNLI数据集来自 - `SNLI Data Set `_ . +以SNLI数据集为例，写一个自定义Pipe的例子如下： - 在 :class:`~fastNLP.io.data_loader.SNLILoader` 的 :meth:`~fastNLP.io.data_loader.SNLILoader._load` - 函数中，我们用以下代码将数据集内容从文本文件读入内存： +.. code-block:: python - .. code-block:: python + from fastNLP.io.loader import SNLILoader + from fastNLP.io.pipe import MatchingPipe - data = SNLILoader().process( - paths='path/to/snli/data', to_lower=False, seq_len_type='seq_len', - get_index=True, concat=False, - ) - print(data) + class MySNLIPipe(MatchingPipe): - 输出的内容是:: + def process(self, data_bundle): + data_bundle = super(MySNLIPipe, self).process(data_bundle) + # MatchingPipe类里封装了一个关于matching任务的process函数，可以直接继承使用 + # 如果有需要进行额外的预处理操作可以在这里加入您的代码 + return data_bundle - In total 3 datasets: - train has 549367 instances. - dev has 9842 instances. - test has 9824 instances. - In total 2 vocabs: - words has 43154 entries. - target has 3 entries. + def process_from_file(self, paths=None): + data_bundle = SNLILoader().load(paths) # 使用SNLILoader读取原始数据集 + # SNLILoader的load函数中，paths如果为None则会自动下载 + return self.process(data_bundle) # 调用相对应的process函数对data_bundle进行处理 +调用Pipe示例： - 这里的data是一个 :class:`~fastNLP.io.base_loader.DataBundle` ，取 ``datasets`` 字典里的内容即可直接传入 - :class:`~fastNLP.Trainer` 或者 :class:`~fastNLP.Tester` 进行训练或者测试。 +.. code-block:: python - :class:`~fastNLP.io.data_loader.IMDBLoader` - 以IMDB数据集为例，在 :class:`~fastNLP.io.data_loader.IMDBLoader` 的 :meth:`~fastNLP.io.data_loader.IMDBLoader._load` - 函数中，我们用以下代码将数据集内容从文本文件读入内存： + from fastNLP.io.pipe import SNLIBertPipe + data_bundle = SNLIBertPipe(lower=True, tokenizer=arg.tokenizer).process_from_file() + print(data_bundle) - .. code-block:: python +输出的内容是:: - data = IMDBLoader().process( - paths={'train': 'path/to/train/file', 'test': 'path/to/test/file'} - ) - print(data) + In total 3 datasets: + train has 549367 instances. + dev has 9842 instances. + test has 9824 instances. + In total 2 vocabs: + words has 34184 entries. + target has 3 entries. - 输出的内容是:: - - In total 3 datasets: - train has 22500 instances. - test has 25000 instances. - dev has 2500 instances. - In total 2 vocabs: - words has 82846 entries. - target has 2 entries. - - - 这里的将原来的train集按9:1的比例分成了训练集和验证集。 +这里表示一共有3个数据集和2个词表。其中： + - 3个数据集分别为train、dev、test数据集，分别有549367、9842、9824个instance + - 2个词表分别为words词表与target词表。其中words词表为句子文本所构建的词表，一共有34184个单词； + target词表为目标标签所构建的词表，一共有3种标签。（注：如果有多个输入，则句子文本所构建的词表将 + 会被命名为words1以对应相对应的列名） ------------------------------------------ -Part V: fastNLP封装好的数据集加载器 +Part V: fastNLP封装好的Loader和Pipe ------------------------------------------ -fastNLP封装好的数据集加载器可以适用于多种类型的任务： - - - `文本分类任务`_ - - `序列标注任务`_ - - `Matching任务`_ - - -文本分类任务 -------------------- - -========================== ================================================================== -数据集名称 数据集加载器 --------------------------- ------------------------------------------------------------------ -IMDb :class:`~fastNLP.io.data_loader.IMDBLoader` --------------------------- ------------------------------------------------------------------ -SST :class:`~fastNLP.io.data_loader.SSTLoader` --------------------------- ------------------------------------------------------------------ -SST-2 :class:`~fastNLP.io.data_loader.SST2Loader` --------------------------- ------------------------------------------------------------------ -Yelp Polarity :class:`~fastNLP.io.data_loader.YelpLoader` --------------------------- ------------------------------------------------------------------ -Yelp Full :class:`~fastNLP.io.data_loader.YelpLoader` --------------------------- ------------------------------------------------------------------ -MTL16 :class:`~fastNLP.io.data_loader.MTL16Loader` -========================== ================================================================== - - - -序列标注任务 -------------------- - -========================== ================================================================== -数据集名称 数据集加载器 --------------------------- ------------------------------------------------------------------ -Conll :class:`~fastNLP.io.data_loader.ConllLoader` --------------------------- ------------------------------------------------------------------ -Conll2003 :class:`~fastNLP.io.data_loader.Conll2003Loader` --------------------------- ------------------------------------------------------------------ -人民日报数据集 :class:`~fastNLP.io.data_loader.PeopleDailyCorpusLoader` -========================== ================================================================== - - - -Matching任务 -------------------- - -========================== ================================================================== -数据集名称 数据集加载器 --------------------------- ------------------------------------------------------------------ -SNLI :class:`~fastNLP.io.data_loader.SNLILoader` --------------------------- ------------------------------------------------------------------ -MultiNLI :class:`~fastNLP.io.data_loader.MNLILoader` --------------------------- ------------------------------------------------------------------ -QNLI :class:`~fastNLP.io.data_loader.QNLILoader` --------------------------- ------------------------------------------------------------------ -RTE :class:`~fastNLP.io.data_loader.RTELoader` --------------------------- ------------------------------------------------------------------ -Quora Pair Dataset :class:`~fastNLP.io.data_loader.QuoraLoader` -========================== ================================================================== +fastNLP封装了多种任务/数据集的Loader和Pipe并提供自动下载功能，具体参见文档 + +`fastNLP可加载的embedding与数据集 `_ diff --git a/docs/source/tutorials/tutorial_3_embedding.rst b/docs/source/tutorials/tutorial_3_embedding.rst index 489b43b4..07dc30bc 100644 --- a/docs/source/tutorials/tutorial_3_embedding.rst +++ b/docs/source/tutorials/tutorial_3_embedding.rst @@ -12,6 +12,7 @@ - `Part IV: 使用预训练的Contextual Embedding(ELMo & BERT)`_ - `Part V: 使用character-level的embedding`_ - `Part VI: 叠加使用多个embedding`_ + - `Part VII: fastNLP支持的预训练Embedding`_ @@ -35,12 +36,14 @@ Part II: 使用随机初始化的embedding .. code-block:: python + from fastNLP import Embedding embed = Embedding(10000, 50) 也可以传入一个初始化的参数矩阵： .. code-block:: python + from fastNLP import Embedding embed = Embedding(init_embed) 其中的init_embed可以是torch.FloatTensor、torch.nn.Embedding或者numpy.ndarray。 @@ -59,6 +62,7 @@ Embedding，例子如下： .. code-block:: python + from fastNLP import StaticEmbedding embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) vocab为根据数据集构建的词表，model_dir_or_name可以是一个路径，也可以是embedding模型的名称： @@ -67,34 +71,13 @@ vocab为根据数据集构建的词表，model_dir_or_name可以是一个路径 和word2vec类型的权重文件都支持) 2 如果传入的是模型名称，那么fastNLP将会根据名称查找embedding模型，如果在cache目录下找到模型则会 - 自动加载；如果找不到则会自动下载。可以通过环境变量 ``FASTNLP_CACHE_DIR`` 来自定义cache目录，如:: + 自动加载；如果找不到则会自动下载到cache目录。默认的cache目录为 `~/.fastNLP` 文件夹。可以通过环境 + 变量 ``FASTNLP_CACHE_DIR`` 来自定义cache目录，如:: $ FASTNLP_CACHE_DIR=~/fastnlp_cache_dir python your_python_file.py 这个命令表示fastNLP将会在 `~/fastnlp_cache_dir` 这个目录下寻找模型，找不到则会自动将模型下载到这个目录 -目前支持的静态embedding模型有： - - ========================== ================================ - 模型名称 模型 - -------------------------- -------------------------------- - en glove.840B.300d - -------------------------- -------------------------------- - en-glove-840d-300 glove.840B.300d - -------------------------- -------------------------------- - en-glove-6b-50 glove.6B.50d - -------------------------- -------------------------------- - en-word2vec-300 谷歌word2vec 300维 - -------------------------- -------------------------------- - en-fasttext 英文fasttext 300维 - -------------------------- -------------------------------- - cn 腾讯中文词向量 200维 - -------------------------- -------------------------------- - cn-fasttext 中文fasttext 300维 - ========================== ================================ - - - ----------------------------------------------------------- Part IV: 使用预训练的Contextual Embedding(ELMo & BERT) ----------------------------------------------------------- @@ -106,62 +89,20 @@ Part IV: 使用预训练的Contextual Embedding(ELMo & BERT) .. code-block:: python + from fastNLP import ElmoEmbedding embed = ElmoEmbedding(vocab, model_dir_or_name='small', requires_grad=False) -目前支持的ElmoEmbedding模型有： - - ========================== ================================ - 模型名称 模型 - -------------------------- -------------------------------- - small allennlp ELMo的small - -------------------------- -------------------------------- - medium allennlp ELMo的medium - -------------------------- -------------------------------- - original allennlp ELMo的original - -------------------------- -------------------------------- - 5.5b-original allennlp ELMo的5.5B original - ========================== ================================ - BERT-embedding的使用方法如下： .. code-block:: python + from fastNLP import BertEmbedding embed = BertEmbedding( vocab, model_dir_or_name='en-base-cased', requires_grad=False, layers='4,-2,-1' ) 其中layers变量表示需要取哪几层的encode结果。 -目前支持的BertEmbedding模型有： - - ========================== ==================================== - 模型名称 模型 - -------------------------- ------------------------------------ - en bert-base-cased - -------------------------- ------------------------------------ - en-base-uncased bert-base-uncased - -------------------------- ------------------------------------ - en-base-cased bert-base-cased - -------------------------- ------------------------------------ - en-large-uncased bert-large-uncased - -------------------------- ------------------------------------ - en-large-cased bert-large-cased - -------------------------- ------------------------------------ - -------------------------- ------------------------------------ - en-large-cased-wwm bert-large-cased-whole-word-mask - -------------------------- ------------------------------------ - en-large-uncased-wwm bert-large-uncased-whole-word-mask - -------------------------- ------------------------------------ - en-base-cased-mrpc bert-base-cased-finetuned-mrpc - -------------------------- ------------------------------------ - -------------------------- ------------------------------------ - multilingual bert-base-multilingual-cased - -------------------------- ------------------------------------ - multilingual-base-uncased bert-base-multilingual-uncased - -------------------------- ------------------------------------ - multilingual-base-cased bert-base-multilingual-cased - ========================== ==================================== - ----------------------------------------------------- Part V: 使用character-level的embedding ----------------------------------------------------- @@ -173,6 +114,7 @@ CNNCharEmbedding的使用例子如下： .. code-block:: python + from fastNLP import CNNCharEmbedding embed = CNNCharEmbedding(vocab, embed_size=100, char_emb_size=50) 这表示这个CNNCharEmbedding当中character的embedding维度大小为50，返回的embedding结果维度大小为100。 @@ -181,12 +123,12 @@ CNNCharEmbedding的使用例子如下： .. code-block:: python + from fastNLP import LSTMCharEmbedding embed = LSTMCharEmbedding(vocab, embed_size=100, char_emb_size=50) 这表示这个LSTMCharEmbedding当中character的embedding维度大小为50，返回的embedding结果维度大小为100。 - ----------------------------------------------------- Part VI: 叠加使用多个embedding ----------------------------------------------------- @@ -197,6 +139,7 @@ Part VI: 叠加使用多个embedding .. code-block:: python + from fastNLP import StaticEmbedding, StackEmbedding embed_1 = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) embed_2 = StaticEmbedding(vocab, model_dir_or_name='en-word2vec-300', requires_grad=True) @@ -208,7 +151,17 @@ StackEmbedding会把多个embedding的结果拼接起来，如上面例子的sta .. code-block:: python + from fastNLP import StaticEmbedding, StackEmbedding, ElmoEmbedding elmo_embedding = ElmoEmbedding(vocab, model_dir_or_name='medium', layers='0,1,2', requires_grad=False) glove_embedding = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) stack_embed = StackEmbedding([elmo_embedding, glove_embedding]) + +------------------------------------------ +Part VII: fastNLP支持的预训练Embedding +------------------------------------------ + +fastNLP支持多种预训练Embedding并提供自动下载功能，具体参见文档 + +`fastNLP可加载的embedding与数据集 `_ + diff --git a/docs/source/tutorials/tutorial_4_loss_optimizer.rst b/docs/source/tutorials/tutorial_4_loss_optimizer.rst index a6e1730a..a53ef89b 100644 --- a/docs/source/tutorials/tutorial_4_loss_optimizer.rst +++ b/docs/source/tutorials/tutorial_4_loss_optimizer.rst @@ -1,4 +1,4 @@ -============================================================================== +============================================================================== 动手实现一个文本分类器I-使用Trainer和Tester快速训练和测试 ============================================================================== @@ -19,7 +19,9 @@ loader = SSTLoader() #这里的all.txt是下载好数据后train.txt、dev.txt、test.txt的组合 - dataset = loader.load("./trainDevTestTrees_PTB/trees/all.txt") + #loader.load(path)会首先判断path是否为none，若是则自动从网站下载数据，若不是则读入数据并返回databundle + databundle_ = loader.load("./trainDevTestTrees_PTB/trees/all.txt") + dataset = databundle_.datasets['train'] print(dataset[0]) 输出数据如下:: @@ -31,6 +33,7 @@ 数据处理 + 可以使用事先定义的 :class:`~fastNLP.io.SSTPipe` 类对数据进行基本预处理，这里我们手动进行处理。 我们使用 :class:`~fastNLP.DataSet` 类的 :meth:`~fastNLP.DataSet.apply` 方法将 ``target`` :mod:`~fastNLP.core.field` 转化为整数。 .. code-block:: python @@ -158,6 +161,7 @@ Vocabulary 的使用 损失函数 训练模型需要提供一个损失函数 ,fastNLP中提供了直接可以导入使用的四种loss，分别为： + * :class:`~fastNLP.CrossEntropyLoss`：包装了torch.nn.functional.cross_entropy()函数，返回交叉熵损失（可以运用于多分类场景） * :class:`~fastNLP.BCELoss`：包装了torch.nn.functional.binary_cross_entropy()函数，返回二分类的交叉熵 * :class:`~fastNLP.L1Loss`：包装了torch.nn.functional.l1_loss()函数，返回L1 损失 @@ -209,7 +213,7 @@ Vocabulary 的使用 #使用CNNText的时候第一个参数输入一个tuple,作为模型定义embedding的参数 #还可以传入 kernel_nums, kernel_sizes, padding, dropout的自定义值 - model_cnn = CNNText((len(vocab),EMBED_DIM), num_classes=3, padding=2, dropout=0.1) + model_cnn = CNNText((len(vocab),EMBED_DIM), num_classes=3, dropout=0.1) #如果在定义trainer的时候没有传入optimizer参数，模型默认的优化器为torch.optim.Adam且learning rate为lr=4e-3 #这里只使用了optimizer_1作为优化器输入，感兴趣可以尝试optimizer_2或者其他优化器作为输入 diff --git a/docs/source/tutorials/tutorial_5_datasetiter.rst b/docs/source/tutorials/tutorial_5_datasetiter.rst index 23d26deb..2ec753c3 100644 --- a/docs/source/tutorials/tutorial_5_datasetiter.rst +++ b/docs/source/tutorials/tutorial_5_datasetiter.rst @@ -20,7 +20,9 @@ loader = SSTLoader() #这里的all.txt是下载好数据后train.txt、dev.txt、test.txt的组合 - dataset = loader.load("./trainDevTestTrees_PTB/trees/all.txt") + #loader.load(path)会首先判断path是否为none，若是则自动从网站下载数据，若不是则读入数据并返回databundle + databundle_ = loader.load("./trainDevTestTrees_PTB/trees/all.txt") + dataset = databundle_.datasets['train'] print(dataset[0]) 输出数据如下:: @@ -32,6 +34,7 @@ 数据处理 + 可以使用事先定义的 :class:`~fastNLP.io.SSTPipe` 类对数据进行基本预处理，这里我们手动进行处理。 我们使用 :class:`~fastNLP.DataSet` 类的 :meth:`~fastNLP.DataSet.apply` 方法将 ``target`` :mod:`~fastNLP.core.field` 转化为整数。 .. code-block:: python @@ -192,7 +195,7 @@ sampler import time embed_dim = 100 - model = CNNText((len(vocab),embed_dim), num_classes=3, padding=2, dropout=0.1) + model = CNNText((len(vocab),embed_dim), num_classes=3, dropout=0.1) def train(epoch, data, devdata): optimizer = torch.optim.Adam(model.parameters(), lr=0.001) diff --git a/docs/source/tutorials/tutorial_6_seq_labeling.rst b/docs/source/tutorials/tutorial_6_seq_labeling.rst index 09a53cdc..7fcf97b3 100644 --- a/docs/source/tutorials/tutorial_6_seq_labeling.rst +++ b/docs/source/tutorials/tutorial_6_seq_labeling.rst @@ -3,64 +3,52 @@ ===================== 这一部分的内容主要展示如何使用fastNLP 实现序列标注任务。你可以使用fastNLP的各个组件快捷，方便地完成序列标注任务，达到出色的效果。 -在阅读这篇Tutorial前，希望你已经熟悉了fastNLP的基础使用，包括基本数据结构以及数据预处理，embedding的嵌入等，希望你对之前的教程有更进一步的掌握。 -我们将对CoNLL-03的英文数据集进行处理，展示如何完成命名实体标注任务整个训练的过程。 +在阅读这篇Tutorial前，希望你已经熟悉了fastNLP的基础使用，尤其是数据的载入以及模型的构建，通过这个小任务的能让你进一步熟悉fastNLP的使用。 +我们将对基于Weibo的中文社交数据集进行处理，展示如何完成命名实体标注任务的整个过程。 载入数据 =================================== -fastNLP可以方便地载入各种类型的数据。同时，针对常见的数据集，我们已经预先实现了载入方法，其中包含CoNLL-03数据集。 +fastNLP的数据载入主要是由Loader与Pipe两个基类衔接完成的。通过Loader可以方便地载入各种类型的数据。同时，针对常见的数据集，我们已经预先实现了载入方法，其中包含weibo数据集。 在设计dataloader时，以DataSetLoader为基类，可以改写并应用于其他数据集的载入。 .. code-block:: python - class Conll2003DataLoader(DataSetLoader): - def __init__(self, task:str='ner', encoding_type:str='bioes'): - assert task in ('ner', 'pos', 'chunk') - index = {'ner':3, 'pos':1, 'chunk':2}[task] - #ConllLoader是fastNLP内置的类 - self._loader = ConllLoader(headers=['raw_words', 'target'], indexes=[0, index]) - self._tag_converters = None - if task in ('ner', 'chunk'): - #iob和iob2bioes会对tag进行统一，标准化 - self._tag_converters = [iob2] - if encoding_type == 'bioes': - self._tag_converters.append(iob2bioes) - - def load(self, path: str): - dataset = self._loader.load(path) - def convert_tag_schema(tags): - for converter in self._tag_converters: - tags = converter(tags) - return tags - if self._tag_converters: - #使用apply实现convert_tag_schema函数，实际上也支持匿名函数 - dataset.apply_field(convert_tag_schema, field_name=Const.TARGET, new_field_name=Const.TARGET) - return dataset - -输出数据格式如： - - {'raw_words': ['on', 'Friday', ':'] type=list, - 'target': ['O', 'O', 'O'] type=list}, + from fastNLP.io import WeiboNERLoader + data_bundle = WeiboNERLoader().load() + + + +载入后的数据如 :: + + {'dev': DataSet( + {{'raw_chars': ['用', '最', '大', '努', '力', '去', '做''人', '生', '。', '哈', '哈', '哈', '哈', '哈', '哈', ' + 'target': ['O', 'O', 'O', 'O', 'O', 'O', 'O', 'O', 'O',, 'O', 'O', 'O', 'O', 'O', 'O'] type=list})} + + {'test': DataSet( + {{'raw_chars': ['感', '恩', '大', '回', '馈'] type=list, 'target': ['O', 'O', 'O', 'O', 'O'] type=list})} + + {'train': DataSet( + {'raw_chars': ['国', '安', '老', '球', '迷'] type=list, 'target': ['B-ORG.NAM', 'I-ORG.NAM', 'B-PER.NOM', 'I-PER.NOM', 'I-PER.NOM'] type=list})} + 数据处理 ---------------------------- -我们进一步处理数据。将数据和词表封装在 :class:`~fastNLP.DataBundle` 类中。data是DataBundle的实例。 -我们输入模型的数据包括char embedding，以及word embedding。在数据处理部分，我们尝试完成词表的构建。 -使用fastNLP中的Vocabulary类来构建词表。 +我们进一步处理数据。通过Pipe基类处理Loader载入的数据。 如果你还有印象，应该还能想起，实现自定义数据集的Pipe时，至少要编写process 函数或者process_from_file 函数。前者接受 :class:`~fastNLP.DataBundle` 类的数据，并返回该 :class:`~fastNLP.DataBundle` 。后者接收数据集所在文件夹为参数，读取并处理为 :class:`~fastNLP.DataBundle` 后，通过process 函数处理数据。 +这里我们已经实现通过Loader载入数据，并已返回 :class:`~fastNLP.DataBundle` 类的数据。我们编写process 函数以处理Loader载入后的数据。 .. code-block:: python - word_vocab = Vocabulary(min_freq=2) - word_vocab.from_dataset(data.datasets['train'], field_name=Const.INPUT) - word_vocab.index_dataset(*data.datasets.values(),field_name=Const.INPUT, new_field_name=Const.INPUT) + from fastNLP.io import ChineseNERPipe + data_bundle = ChineseNERPipe(encoding_type='bioes', bigram=True).process(data_bundle) -处理后的data对象内部为： +载入后的数据如下 :: - dataset - vocabs - dataset保存了train和test中的数据，并保存为dataset类型 - vocab保存了words，raw-words以及target的词表。 + {'raw_chars': ['用', '最', '大', '努', '力', '去', '做', '值', '得', '的', '事', '人', '生', '。', '哈', '哈', '哈', '哈', '哈', '哈', '我', '在'] type=list, + 'target': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] type=list, + 'chars': [97, 71, 34, 422, 104, 72, 144, 628, 66, 3, 158, 2, 9, 647, 485, 196, 2,19] type=list, + 'bigrams': [5948, 1950, 34840, 98, 8413, 3961, 34841, 631, 34842, 407, 462, 45, 3 1959, 1619, 3, 3, 3, 3, 3, 2663, 29, 90] type=list, + 'seq_len': 30 type=int} 模型构建 -------------------------------- @@ -69,27 +57,23 @@ fastNLP可以方便地载入各种类型的数据。同时，针对常见的数 模型的训练 首先实例化模型，导入所需的char embedding以及word embedding。Embedding的载入可以参考教程。 -也可以查看 :mod:`~fastNLP.modules.encoder.embedding` 使用所需的embedding 载入方法。 -fastNLP将模型的训练过程封装在了 :class:`~fastnlp.trainer` 类中。 +也可以查看 :mod:`~fastNLP.embedding` 使用所需的embedding 载入方法。 +fastNLP将模型的训练过程封装在了 :class:`~fastnlp.Trainer` 类中。 根据不同的任务调整trainer中的参数即可。通常，一个trainer实例需要有：指定的训练数据集，模型，优化器，loss函数，评测指标，以及指定训练的epoch数，batch size等参数。 .. code-block:: python #实例化模型 - model = CNNBiLSTMCRF(word_embed, char_embed, hidden_size=200, num_layers=1, tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type) - #定义优化器 - optimizer = Adam(model.parameters(), lr=0.005) + model = CNBiLSTMCRFNER(char_embed, num_classes=len(data_bundle.vocabs['target']), bigram_embed=bigram_embed) #定义评估指标 - Metrics=SpanFPreRecMetric(tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type) - #实例化trainer - trainer = Trainer(train_data=data.datasets['train'], model=model, optimizer=optimizer, dev_data=data.datasets['test'], batch_size=10, metrics=Metrics,callbacks=callbacks, n_epochs=100) - #开始训练 - trainer.train() + Metrics=SpanFPreRecMetric(data_bundle.vocabs['target'], encoding_type='bioes') + #实例化trainer并训练 + Trainer(data_bundle.datasets['train'], model, batch_size=20, metrics=Metrics, num_workers=2, dev_data=data_bundle. datasets['dev']).train() + 训练中会保存最优的参数配置。 -训练的结果如下： -.. code-block:: python +训练的结果如下 :: Evaluation on DataSet test: SpanFPreRecMetric: f=0.727661, pre=0.732293, rec=0.723088 diff --git a/docs/source/tutorials/tutorial_9_callback.rst b/docs/source/tutorials/tutorial_9_callback.rst index 8e2742bb..dc50aca5 100644 --- a/docs/source/tutorials/tutorial_9_callback.rst +++ b/docs/source/tutorials/tutorial_9_callback.rst @@ -23,7 +23,7 @@ Callback的构建和使用 class LRDecay(fastNLP.Callback): def __init__(self): - super(MyCallback, self).__init__() + super(LRDecay, self).__init__() self.base_lrs = [] self.delta = [] diff --git a/docs/source/user/tutorials.rst b/docs/source/user/tutorials.rst index 196f9c29..3e9e1b54 100644 --- a/docs/source/user/tutorials.rst +++ b/docs/source/user/tutorials.rst @@ -8,7 +8,7 @@ fastNLP 详细使用教程 :maxdepth: 1 使用DataSet预处理文本 - 使用DataSetLoader加载数据集 + 使用Loader和Pipe加载并处理数据集 使用Embedding模块将文本转成向量 动手实现一个文本分类器I-使用Trainer和Tester快速训练和测试 动手实现一个文本分类器II-使用DataSetIter实现自定义训练过程 diff --git a/fastNLP/__init__.py b/fastNLP/__init__.py index ec192568..aceaf47f 100644 --- a/fastNLP/__init__.py +++ b/fastNLP/__init__.py @@ -13,11 +13,12 @@ fastNLP 中最常用的组件可以直接从 fastNLP 包中 import ，他们的 __all__ = [ "Instance", "FieldArray", - + + "DataSetIter", "BatchIter", "TorchLoaderIter", - + "Vocabulary", "DataSet", "Const", @@ -31,6 +32,7 @@ __all__ = [ "TensorboardCallback", "LRScheduler", "ControlC", + "LRFinder", "Padder", "AutoPadder", @@ -43,7 +45,8 @@ __all__ = [ "Optimizer", "SGD", "Adam", - + "AdamW", + "Sampler", "SequentialSampler", "BucketSampler", @@ -51,16 +54,23 @@ __all__ = [ "LossFunc", "CrossEntropyLoss", - "L1Loss", "BCELoss", + "L1Loss", + "BCELoss", "NLLLoss", "LossInForward", - "cache_results" + "cache_results", + + 'logger' ] __version__ = '0.4.5' -from .core import * +from . import embeddings from . import models from . import modules -from . import embeddings -from .io import data_loader +from .core import * +from .io import loader, pipe + +import sys +from .doc_utils import doc_process +doc_process(sys.modules[__name__]) \ No newline at end of file diff --git a/fastNLP/core/__init__.py b/fastNLP/core/__init__.py index c9f51123..efee08b5 100644 --- a/fastNLP/core/__init__.py +++ b/fastNLP/core/__init__.py @@ -10,21 +10,85 @@ core 模块里实现了 fastNLP 的核心框架，常用的功能都可以从 fa 对于常用的功能，你只需要在 :doc:`fastNLP` 中查看即可。如果想了解各个子模块的具体作用，您可以在下面找到每个子模块的具体文档。 -.. todo:: - 介绍core 的子模块的分工，好像必要性不大 - """ +__all__ = [ + "DataSet", + + "Instance", + + "FieldArray", + "Padder", + "AutoPadder", + "EngChar2DPadder", + + "Vocabulary", + + "DataSetIter", + "BatchIter", + "TorchLoaderIter", + + "Const", + + "Tester", + "Trainer", + + "cache_results", + "seq_len_to_mask", + "get_seq_len", + "logger", + + "Callback", + "GradientClipCallback", + "EarlyStopCallback", + "FitlogCallback", + "EvaluateCallback", + "LRScheduler", + "ControlC", + "LRFinder", + "TensorboardCallback", + "WarmupCallback", + 'SaveModelCallback', + "EchoCallback", + "TesterCallback", + "CallbackException", + "EarlyStopError", + + "LossFunc", + "CrossEntropyLoss", + "L1Loss", + "BCELoss", + "NLLLoss", + "LossInForward", + + "AccuracyMetric", + "SpanFPreRecMetric", + "ExtractiveQAMetric", + + "Optimizer", + "SGD", + "Adam", + "AdamW", + + "SequentialSampler", + "BucketSampler", + "RandomSampler", + "Sampler", +] + +from ._logger import logger from .batch import DataSetIter, BatchIter, TorchLoaderIter -from .callback import Callback, GradientClipCallback, EarlyStopCallback, TensorboardCallback, LRScheduler, ControlC +from .callback import Callback, GradientClipCallback, EarlyStopCallback, FitlogCallback, EvaluateCallback, \ + LRScheduler, ControlC, LRFinder, TensorboardCallback, WarmupCallback, SaveModelCallback, EchoCallback, \ + TesterCallback, CallbackException, EarlyStopError from .const import Const from .dataset import DataSet from .field import FieldArray, Padder, AutoPadder, EngChar2DPadder from .instance import Instance from .losses import LossFunc, CrossEntropyLoss, L1Loss, BCELoss, NLLLoss, LossInForward from .metrics import AccuracyMetric, SpanFPreRecMetric, ExtractiveQAMetric -from .optimizer import Optimizer, SGD, Adam +from .optimizer import Optimizer, SGD, Adam, AdamW from .sampler import SequentialSampler, BucketSampler, RandomSampler, Sampler from .tester import Tester from .trainer import Trainer -from .utils import cache_results, seq_len_to_mask +from .utils import cache_results, seq_len_to_mask, get_seq_len from .vocabulary import Vocabulary diff --git a/fastNLP/core/_logger.py b/fastNLP/core/_logger.py new file mode 100644 index 00000000..7198cfbd --- /dev/null +++ b/fastNLP/core/_logger.py @@ -0,0 +1,155 @@ +"""undocumented""" + +__all__ = [ + 'logger', +] + +import logging +import logging.config +import os +import sys +import warnings + +ROOT_NAME = 'fastNLP' + +try: + import fitlog +except ImportError: + fitlog = None +try: + from tqdm.auto import tqdm +except ImportError: + tqdm = None + +if tqdm is not None: + class TqdmLoggingHandler(logging.Handler): + def __init__(self, level=logging.INFO): + super().__init__(level) + + def emit(self, record): + try: + msg = self.format(record) + tqdm.write(msg) + self.flush() + except (KeyboardInterrupt, SystemExit): + raise + except: + self.handleError(record) +else: + class TqdmLoggingHandler(logging.StreamHandler): + def __init__(self, level=logging.INFO): + super().__init__(sys.stdout) + self.setLevel(level) + + +def _get_level(level): + if isinstance(level, int): + pass + else: + level = level.lower() + level = {'info': logging.INFO, 'debug': logging.DEBUG, + 'warn': logging.WARN, 'warning': logging.WARN, + 'error': logging.ERROR}[level] + return level + + +def _add_file_handler(logger, path, level='INFO'): + for h in logger.handlers: + if isinstance(h, logging.FileHandler): + if os.path.abspath(path) == h.baseFilename: + # file path already added + return + + # File Handler + if os.path.exists(path): + assert os.path.isfile(path) + warnings.warn('log already exists in {}'.format(path)) + dirname = os.path.abspath(os.path.dirname(path)) + os.makedirs(dirname, exist_ok=True) + + file_handler = logging.FileHandler(path, mode='a') + file_handler.setLevel(_get_level(level)) + file_formatter = logging.Formatter(fmt='%(asctime)s - %(module)s - [%(levelname)s] - %(message)s', + datefmt='%Y/%m/%d %H:%M:%S') + file_handler.setFormatter(file_formatter) + logger.addHandler(file_handler) + + +def _set_stdout_handler(logger, stdout='tqdm', level='INFO'): + level = _get_level(level) + if stdout not in ['none', 'plain', 'tqdm']: + raise ValueError('stdout must in one of {}'.format(['none', 'plain', 'tqdm'])) + # make sure to initialize logger only once + stream_handler = None + for i, h in enumerate(logger.handlers): + if isinstance(h, (logging.StreamHandler, TqdmLoggingHandler)): + stream_handler = h + break + if stream_handler is not None: + logger.removeHandler(stream_handler) + + # Stream Handler + if stdout == 'plain': + stream_handler = logging.StreamHandler(sys.stdout) + elif stdout == 'tqdm': + stream_handler = TqdmLoggingHandler(level) + else: + stream_handler = None + + if stream_handler is not None: + stream_formatter = logging.Formatter('%(message)s') + stream_handler.setLevel(level) + stream_handler.setFormatter(stream_formatter) + logger.addHandler(stream_handler) + + +class FastNLPLogger(logging.getLoggerClass()): + def __init__(self, name): + super().__init__(name) + + def add_file(self, path='./log.txt', level='INFO'): + """add log output file and level""" + _add_file_handler(self, path, level) + + def set_stdout(self, stdout='tqdm', level='INFO'): + """set stdout format and level""" + _set_stdout_handler(self, stdout, level) + + +logging.setLoggerClass(FastNLPLogger) + + +# print(logging.getLoggerClass()) +# print(logging.getLogger()) + +def _init_logger(path=None, stdout='tqdm', level='INFO'): + """initialize logger""" + level = _get_level(level) + + # logger = logging.getLogger() + logger = logging.getLogger(ROOT_NAME) + logger.propagate = False + logger.setLevel(level) + + _set_stdout_handler(logger, stdout, level) + + # File Handler + if path is not None: + _add_file_handler(logger, path, level) + + return logger + + +def _get_logger(name=None, level='INFO'): + level = _get_level(level) + if name is None: + name = ROOT_NAME + assert isinstance(name, str) + if not name.startswith(ROOT_NAME): + name = '{}.{}'.format(ROOT_NAME, name) + logger = logging.getLogger(name) + logger.setLevel(level) + return logger + + +logger = _init_logger(path=None) diff --git a/fastNLP/core/_parallel_utils.py b/fastNLP/core/_parallel_utils.py index 4a7757d3..ce745820 100644 --- a/fastNLP/core/_parallel_utils.py +++ b/fastNLP/core/_parallel_utils.py @@ -1,10 +1,14 @@ +"""undocumented""" + +__all__ = [] import threading + import torch +from torch import nn from torch.nn.parallel.parallel_apply import get_a_var - -from torch.nn.parallel.scatter_gather import scatter_kwargs, gather from torch.nn.parallel.replicate import replicate +from torch.nn.parallel.scatter_gather import scatter_kwargs, gather def parallel_apply(modules, func_name, inputs, kwargs_tup=None, devices=None): @@ -26,11 +30,11 @@ def parallel_apply(modules, func_name, inputs, kwargs_tup=None, devices=None): assert len(modules) == len(devices) else: devices = [None] * len(modules) - + lock = threading.Lock() results = {} grad_enabled = torch.is_grad_enabled() - + def _worker(i, module, input, kwargs, device=None): torch.set_grad_enabled(grad_enabled) if device is None: @@ -46,20 +50,20 @@ def parallel_apply(modules, func_name, inputs, kwargs_tup=None, devices=None): except Exception as e: with lock: results[i] = e - + if len(modules) > 1: threads = [threading.Thread(target=_worker, args=(i, module, input, kwargs, device)) for i, (module, input, kwargs, device) in enumerate(zip(modules, inputs, kwargs_tup, devices))] - + for thread in threads: thread.start() for thread in threads: thread.join() else: _worker(0, modules[0], inputs[0], kwargs_tup[0], devices[0]) - + outputs = [] for i in range(len(inputs)): output = results[i] @@ -78,6 +82,7 @@ def _data_parallel_wrapper(func_name, device_ids, output_device): :param output_device: nn.DataParallel中的output_device :return: """ + def wrapper(network, *inputs, **kwargs): inputs, kwargs = scatter_kwargs(inputs, kwargs, device_ids, dim=0) if len(device_ids) == 1: @@ -85,4 +90,18 @@ def _data_parallel_wrapper(func_name, device_ids, output_device): replicas = replicate(network, device_ids[:len(inputs)]) outputs = parallel_apply(replicas, func_name, inputs, kwargs, device_ids[:len(replicas)]) return gather(outputs, output_device) + return wrapper + + +def _model_contains_inner_module(model): + """ + + :param nn.Module model: 模型文件，判断是否内部包含model.module, 多用于check模型是否是nn.DataParallel, + nn.parallel.DistributedDataParallel。主要是在做形参匹配的时候需要使用最内部的model的function。 + :return: bool + """ + if isinstance(model, nn.Module): + if isinstance(model, (nn.DataParallel, nn.parallel.DistributedDataParallel)): + return True + return False diff --git a/fastNLP/core/batch.py b/fastNLP/core/batch.py index 64c5f48e..b14b21de 100644 --- a/fastNLP/core/batch.py +++ b/fastNLP/core/batch.py @@ -9,14 +9,15 @@ __all__ = [ ] import atexit +from numbers import Number import numpy as np import torch import torch.utils.data -from numbers import Number -from .sampler import SequentialSampler +from ._logger import logger from .dataset import DataSet +from .sampler import SequentialSampler _python_is_exit = False @@ -48,6 +49,11 @@ class DataSetGetter: return len(self.dataset) def collate_fn(self, batch: list): + """ + + :param batch: [[idx1, x_dict1, y_dict1], [idx2, x_dict2, y_dict2], [xx, xx, xx]] + :return: + """ # TODO 支持在DataSet中定义collate_fn，因为有时候可能需要不同的field之间融合，比如BERT的场景 batch_x = {n:[] for n in self.inputs.keys()} batch_y = {n:[] for n in self.targets.keys()} @@ -70,7 +76,7 @@ class DataSetGetter: try: data, flag = _to_tensor(data, f.dtype) except TypeError as e: - print(f"Field {n} cannot be converted to torch.tensor.") + logger.error(f"Field {n} cannot be converted to torch.tensor.") raise e batch_dict[n] = data return batch_dict @@ -93,9 +99,13 @@ class DataSetGetter: class SamplerAdapter(torch.utils.data.Sampler): def __init__(self, sampler, dataset): + super().__init__(dataset) self.sampler = sampler self.dataset = dataset + def __len__(self): + return len(self.dataset) + def __iter__(self): return iter(self.sampler(self.dataset)) @@ -136,8 +146,6 @@ class BatchIter: class DataSetIter(BatchIter): """ - 别名：:class:`fastNLP.DataSetIter` :class:`fastNLP.core.batch.DataSetIter` - DataSetIter 用于从 `DataSet` 中按一定的顺序, 依次按 ``batch_size`` 的大小将数据取出， 组成 `x` 和 `y`:: @@ -146,34 +154,41 @@ class DataSetIter(BatchIter): for batch_x, batch_y in batch: # do stuff ... - :param dataset: :class:`~fastNLP.DataSet` 对象, 数据集 - :param int batch_size: 取出的batch大小 - :param sampler: 规定使用的 :class:`~fastNLP.Sampler` 方式. 若为 ``None`` , 使用 :class:`~fastNLP.SequentialSampler`. - - Default: ``None`` - :param bool as_numpy: 若为 ``True`` , 输出batch为 numpy.array. 否则为 :class:`torch.Tensor`. - - Default: ``False`` - :param int num_workers: 使用多少个进程来预处理数据 - :param bool pin_memory: 是否将产生的tensor使用pin memory, 可能会加快速度。 - :param bool drop_last: 如果最后一个batch没有batch_size这么多sample，就扔掉最后一个 - :param timeout: - :param worker_init_fn: 在每个worker启动时调用该函数，会传入一个值，该值是worker的index。 """ def __init__(self, dataset, batch_size=1, sampler=None, as_numpy=False, num_workers=0, pin_memory=False, drop_last=False, timeout=0, worker_init_fn=None): + """ + + :param dataset: :class:`~fastNLP.DataSet` 对象, 数据集 + :param int batch_size: 取出的batch大小 + :param sampler: 规定使用的 :class:`~fastNLP.Sampler` 方式. 若为 ``None`` , 使用 :class:`~fastNLP.SequentialSampler`. + + Default: ``None`` + :param bool as_numpy: 若为 ``True`` , 输出batch为 numpy.array. 否则为 :class:`torch.Tensor`. + + Default: ``False`` + :param int num_workers: 使用多少个进程来预处理数据 + :param bool pin_memory: 是否将产生的tensor使用pin memory, 可能会加快速度。 + :param bool drop_last: 如果最后一个batch没有batch_size这么多sample，就扔掉最后一个 + :param timeout: + :param worker_init_fn: 在每个worker启动时调用该函数，会传入一个值，该值是worker的index。 + """ super().__init__() assert isinstance(dataset, DataSet) - sampler = SamplerAdapter(sampler=sampler or SequentialSampler(), dataset=dataset) + if not isinstance(sampler, torch.utils.data.Sampler): + self.sampler = SamplerAdapter(sampler=sampler or SequentialSampler(), dataset=dataset) + else: + self.sampler = sampler dataset = DataSetGetter(dataset, as_numpy) collate_fn = dataset.collate_fn if hasattr(dataset, 'collate_fn') else None self.dataiter = torch.utils.data.DataLoader( - dataset=dataset, batch_size=batch_size, sampler=sampler, + dataset=dataset, batch_size=batch_size, sampler=self.sampler, collate_fn=collate_fn, num_workers=num_workers, pin_memory=pin_memory, drop_last=drop_last, timeout=timeout, worker_init_fn=worker_init_fn) - self.num_batches = self.get_num_batches(len(dataset), batch_size, drop_last) + # 以sampler的数量为准，因为DistributedSampler的时候每个进程上并不是所有的数据都用上了 + self.num_batches = self.get_num_batches(len(self.dataiter.sampler), batch_size, drop_last) self.batch_size = batch_size @@ -182,7 +197,7 @@ class TorchLoaderIter(BatchIter): super().__init__() assert isinstance(dataset, torch.utils.data.DataLoader) self.dataiter = dataset - self.num_batches = self.get_num_batches(len(dataset), dataset.batch_size, dataset.drop_last) + self.num_batches = self.get_num_batches(len(dataset.sampler), dataset.batch_size, dataset.drop_last) self.batch_size = dataset.batch_size @@ -200,6 +215,13 @@ class OnlineDataIter(BatchIter): def _to_tensor(batch, field_dtype): + """ + + :param batch: np.array() + :param field_dtype: 数据类型 + :return: batch, flag. 如果传入的数据支持转为tensor，返回的batch就是tensor，且flag为True；如果传入的数据不支持转为tensor， + 返回的batch就是原来的数据，且flag为False + """ try: if field_dtype is not None and isinstance(field_dtype, type)\ and issubclass(field_dtype, Number) \ diff --git a/fastNLP/core/callback.py b/fastNLP/core/callback.py index 6f855397..fe198acc 100644 --- a/fastNLP/core/callback.py +++ b/fastNLP/core/callback.py @@ -51,22 +51,30 @@ callback模块实现了 fastNLP 中的许多 callback 类，用于增强 :class: """ __all__ = [ "Callback", + "GradientClipCallback", "EarlyStopCallback", - "TensorboardCallback", "FitlogCallback", + "EvaluateCallback", "LRScheduler", "ControlC", + "LRFinder", + "TensorboardCallback", + "WarmupCallback", + "SaveModelCallback", + "EchoCallback", + "TesterCallback", "CallbackException", "EarlyStopError" ] import os +import sys +from copy import deepcopy import torch -from copy import deepcopy -import sys + from .utils import _save_model try: @@ -76,9 +84,9 @@ try: except: tensorboardX_flag = False -from ..io.model_io import ModelSaver, ModelLoader from .dataset import DataSet from .tester import Tester +from ._logger import logger try: import fitlog @@ -88,8 +96,6 @@ except: class Callback(object): """ - 别名：:class:`fastNLP.Callback` :class:`fastNLP.core.callback.Callback` - Callback是fastNLP中被设计用于增强 :class:`~fastNLP.Trainer` 的类。 如果Callback被传递给了 Trainer , 则 Trainer 会在对应的阶段调用Callback的函数， 具体调用时机可以通过 :doc:`trainer 模块` 查看。 @@ -100,7 +106,8 @@ class Callback(object): def __init__(self): super(Callback, self).__init__() self._trainer = None # 在Trainer内部被重新赋值 - + self._disabled = False + @property def trainer(self): """ @@ -158,7 +165,19 @@ class Callback(object): def batch_per_epoch(self): """每个epoch一共有多少个batch，只有在on_epoch_begin之后才能调用该属性。""" return self._trainer.batch_per_epoch - + + @property + def is_master(self): + return self._trainer.is_master + + @property + def disabled(self): + return self._disabled + + @property + def logger(self): + return getattr(self._trainer, 'logger', logger) + def on_train_begin(self): """ 在Train过程开始之前调用。 @@ -250,6 +269,14 @@ class Callback(object): :return: """ pass + + def on_validation(self): + """ + 如果Trainer中设置了验证，则会在每次需要验证时调用该函数 + + :return: + """ + pass def on_epoch_end(self): """ @@ -281,6 +308,8 @@ def _transfer(func): def wrapper(manager, *arg): returns = [] for callback in manager.callbacks: + if callback.disabled: + continue returns.append(getattr(callback, func.__name__)(*arg)) return returns @@ -288,31 +317,39 @@ def _transfer(func): class CallbackManager(Callback): + """ + 内部使用的Callback管理类 + """ def __init__(self, env, callbacks=None): """ - 内部使用的Callback管理类 :param dict env: The key is the name of the Trainer attribute(str). The value is the attribute itself. :param List[Callback] callbacks: """ super(CallbackManager, self).__init__() # set attribute of trainer environment - + self._env = env self.callbacks = [] - if callbacks is not None: - if isinstance(callbacks, list): - if all([isinstance(cb, Callback) for cb in callbacks]) is True: - self.callbacks.extend(callbacks) - else: - obj = [not isinstance(cb, Callback) for cb in callbacks][0] - raise TypeError(f"Expect sub-classes of Callback. Got {type(obj)}") + if callbacks: + self.callbacks = self.prepare_callbacks(callbacks) + + def prepare_callbacks(self, callbacks): + if not callbacks: + return [] + if isinstance(callbacks, list): + if all([isinstance(cb, Callback) for cb in callbacks]) is True: + pass else: - raise TypeError(f"Expect callbacks in CallbackManager(callbacks) to be list. Got {type(callbacks)}.") - - for env_name, env_val in env.items(): - for callback in self.callbacks: + obj = [not isinstance(cb, Callback) for cb in callbacks][0] + raise TypeError(f"Expect sub-classes of Callback. Got {type(obj)}") + else: + raise TypeError(f"Expect callbacks in CallbackManager(callbacks) to be list. Got {type(callbacks)}.") + + for env_name, env_val in self._env.items(): + for callback in callbacks: setattr(callback, '_' + env_name, env_val) # Callback.trainer - + return callbacks + @_transfer def on_train_begin(self): pass @@ -352,6 +389,10 @@ class CallbackManager(Callback): @_transfer def on_valid_end(self, eval_result, metric_key, optimizer, is_better_eval): pass + + @_transfer + def on_validation(self): + pass @_transfer def on_epoch_end(self): @@ -366,28 +407,53 @@ class CallbackManager(Callback): pass +class DistCallbackManager(CallbackManager): + def __init__(self, env, callbacks_all=None, callbacks_master=None): + super(DistCallbackManager, self).__init__(env) + assert 'trainer' in env + self._trainer = env['trainer'] + self.callbacks_master = [] + self.callbacks_all = [] + self.add_callback(callbacks_all, master=False) + self.add_callback(callbacks_master, master=True) + + def patch_callback(self, callbacks, disabled): + if not callbacks: + return + if not isinstance(callbacks, (list, tuple)): + callbacks = [callbacks] + for cb in callbacks: + cb._disabled = disabled + + def add_callback(self, cb, master=False): + if master: + self.patch_callback(cb, not self.is_master) + self.callbacks_master += self.prepare_callbacks(cb) + else: + self.callbacks_all += self.prepare_callbacks(cb) + self.callbacks = self.callbacks_all + self.callbacks_master + + class GradientClipCallback(Callback): """ - 别名：:class:`fastNLP.GradientClipCallback` :class:`fastNLP.core.callback.GradientClipCallback` - 每次backward前，将parameter的gradient clip到某个范围。 - - :param None,torch.Tensor,List[torch.Tensor] parameters: 一般通过model.parameters()获得。 - 如果为None则默认对Trainer的model中所有参数进行clip - :param float clip_value: 将gradient 限制到[-clip_value, clip_value]。clip_value应该为正数 - :param str clip_type: 支持'norm', 'value' - 两种:: - - 1 'norm', 将gradient的norm rescale到[-clip_value, clip_value] - - 2 'value', 将gradient限制在[-clip_value, clip_value], - 小于-clip_value的gradient被赋值为-clip_value; - 大于clip_value的gradient被赋值为clip_value. - """ def __init__(self, parameters=None, clip_value=1, clip_type='norm'): + """ + :param None,torch.Tensor,List[torch.Tensor] parameters: 一般通过model.parameters()获得。 + 如果为None则默认对Trainer的model中所有参数进行clip + :param float clip_value: 将gradient 限制到[-clip_value, clip_value]。clip_value应该为正数 + :param str clip_type: 支持'norm', 'value' + 两种:: + + 1 'norm', 将gradient的norm rescale到[-clip_value, clip_value] + + 2 'value', 将gradient限制在[-clip_value, clip_value], + 小于-clip_value的gradient被赋值为-clip_value; + 大于clip_value的gradient被赋值为clip_value. + """ super().__init__() from torch import nn @@ -403,6 +469,9 @@ class GradientClipCallback(Callback): def on_backward_end(self): if self.step%self.update_every==0: if self.parameters is None: + if getattr(self.trainer, 'fp16', ''): + from apex import amp + self.clip_fun(amp.master_params(self.optimizer), self.clip_value) self.clip_fun(self.model.parameters(), self.clip_value) else: self.clip_fun(self.parameters, self.clip_value) @@ -410,14 +479,14 @@ class GradientClipCallback(Callback): class EarlyStopCallback(Callback): """ - 别名：:class:`fastNLP.EarlyStopCallback` :class:`fastNLP.core.callback.EarlyStopCallback` - - 多少个epoch没有变好就停止训练，相关类 :class:`EarlyStopError` - - :param int patience: epoch的数量 + 多少个epoch没有变好就停止训练，相关类 :class:`~fastNLP.core.callback.EarlyStopError` """ def __init__(self, patience): + """ + + :param int patience: epoch的数量 + """ super(EarlyStopCallback, self).__init__() self.patience = patience self.wait = 0 @@ -434,52 +503,54 @@ class EarlyStopCallback(Callback): def on_exception(self, exception): if isinstance(exception, EarlyStopError): - print("Early Stopping triggered in epoch {}!".format(self.epoch)) + logger.info("Early Stopping triggered in epoch {}!".format(self.epoch)) else: raise exception # 抛出陌生Error class FitlogCallback(Callback): """ - 别名: :class:`fastNLP.FitlogCallback` :class:`fastNLP.core.callback.FitlogCallback` - 该callback可将loss和progress写入到fitlog中; 如果Trainer有dev的数据，将自动把dev的结果写入到log中; 同时还支持传入 - 一个(或多个)test数据集进行测试(只有在trainer具有dev时才能使用)，每次在dev上evaluate之后会在这些数据集上验证一下。 - 并将验证结果写入到fitlog中。这些数据集的结果是根据dev上最好的结果报道的，即如果dev在第3个epoch取得了最佳，则 - fitlog中记录的关于这些数据集的结果就是来自第三个epoch的结果。 - - :param ~fastNLP.DataSet,Dict[~fastNLP.DataSet] data: 传入DataSet对象，会使用多个Trainer中的metric对数据进行验证。如果需要传入多个 - DataSet请通过dict的方式传入，dict的key将作为对应dataset的name传递给fitlog。若tester不为None时，data需要通过 - dict的方式传入。如果仅传入DataSet, 则被命名为test - :param ~fastNLP.Tester tester: Tester对象，将在on_valid_end时调用。tester中的DataSet会被称为为`test` - :param int log_loss_every: 多少个step记录一次loss(记录的是这几个batch的loss平均值)，如果数据集较大建议将该值设置得 - 大一些，不然会导致log文件巨大。默认为0, 即不要记录loss。 - :param int verbose: 是否在终端打印evaluation的结果，0不打印。 - :param bool log_exception: fitlog是否记录发生的exception信息 + 一个(或多个)test数据集进行测试(只有在trainer具有dev时才能使用)，每次在dev上evaluate之后会在这些数据集上验证一下。 + 并将验证结果写入到fitlog中。这些数据集的结果是根据dev上最好的结果报道的，即如果dev在第3个epoch取得了最佳，则 + fitlog中记录的关于这些数据集的结果就是来自第三个epoch的结果。 """ def __init__(self, data=None, tester=None, log_loss_every=0, verbose=0, log_exception=False): + """ + + :param ~fastNLP.DataSet,Dict[~fastNLP.DataSet] data: 传入DataSet对象，会使用多个Trainer中的metric对数据进行验证。如果需要 + 传入多个DataSet请通过dict的方式传入，dict的key将作为对应dataset的name传递给fitlog。data的结果的名称以'data'开头。 + :param ~fastNLP.Tester,Dict[~fastNLP.Tester] tester: Tester对象，将在on_valid_end时调用。tester的结果的名称以'tester'开头 + :param int log_loss_every: 多少个step记录一次loss(记录的是这几个batch的loss平均值)，如果数据集较大建议将该值设置得 + 大一些，不然会导致log文件巨大。默认为0, 即不要记录loss。 + :param int verbose: 是否在终端打印evaluation的结果，0不打印。 + :param bool log_exception: fitlog是否记录发生的exception信息 + """ super().__init__() self.datasets = {} self.testers = {} self._log_exception = log_exception assert isinstance(log_loss_every, int) and log_loss_every>=0 if tester is not None: - assert isinstance(tester, Tester), "Only fastNLP.Tester allowed." - assert isinstance(data, dict) or data is None, "If tester is not None, only dict[DataSet] allowed for data." - if data is not None: - assert 'test' not in data, "Cannot use `test` as DataSet key, when tester is passed." - setattr(tester, 'verbose', 0) - self.testers['test'] = tester - + if isinstance(tester, dict): + for name, test in tester.items(): + if not isinstance(test, Tester): + raise TypeError(f"{name} in tester is not a valid fastNLP.Tester.") + self.testers['tester-' + name] = test + if isinstance(tester, Tester): + self.testers['tester-test'] = tester + for tester in self.testers.values(): + setattr(tester, 'verbose', 0) + if isinstance(data, dict): for key, value in data.items(): assert isinstance(value, DataSet), f"Only DataSet object is allowed, not {type(value)}." for key, value in data.items(): - self.datasets[key] = value + self.datasets['data-' + key] = value elif isinstance(data, DataSet): - self.datasets['test'] = data - else: + self.datasets['data-test'] = data + elif data is not None: raise TypeError("data receives dict[DataSet] or DataSet object.") self.verbose = verbose @@ -492,8 +563,11 @@ class FitlogCallback(Callback): if len(self.datasets) > 0: for key, data in self.datasets.items(): - tester = Tester(data=data, model=self.model, batch_size=self.batch_size, metrics=self.trainer.metrics, - verbose=0) + tester = Tester(data=data, model=self.model, + batch_size=self.trainer.kwargs.get('dev_batch_size', self.batch_size), + metrics=self.trainer.metrics, + verbose=0, + use_tqdm=self.trainer.test_use_tqdm) self.testers[key] = tester fitlog.add_progress(total_steps=self.n_steps) @@ -533,17 +607,76 @@ class FitlogCallback(Callback): fitlog.add_other(repr(exception), name='except_info') -class LRScheduler(Callback): +class EvaluateCallback(Callback): + """ + 该callback用于扩展Trainer训练过程中只能对dev数据进行验证的问题。 """ - 别名：:class:`fastNLP.LRScheduler` :class:`fastNLP.core.callback.LRScheduler` - 对PyTorch LR Scheduler的包装以使得其可以被Trainer所使用 + def __init__(self, data=None, tester=None): + """ + :param ~fastNLP.DataSet,Dict[~fastNLP.DataSet] data: 传入DataSet对象，会使用多个Trainer中的metric对数据进行验证。如果需要传入多个 + DataSet请通过dict的方式传入。 + :param ~fastNLP.Tester,Dict[~fastNLP.DataSet] tester: Tester对象，将在on_valid_end时调用。 + """ + super().__init__() + self.datasets = {} + self.testers = {} + if tester is not None: + if isinstance(tester, dict): + for name, test in tester.items(): + if not isinstance(test, Tester): + raise TypeError(f"{name} in tester is not a valid fastNLP.Tester.") + self.testers['tester-' + name] = test + if isinstance(tester, Tester): + self.testers['tester-test'] = tester + for tester in self.testers.values(): + setattr(tester, 'verbose', 0) + + if isinstance(data, dict): + for key, value in data.items(): + assert isinstance(value, DataSet), f"Only DataSet object is allowed, not {type(value)}." + for key, value in data.items(): + self.datasets['data-' + key] = value + elif isinstance(data, DataSet): + self.datasets['data-test'] = data + elif data is not None: + raise TypeError("data receives dict[DataSet] or DataSet object.") + + def on_train_begin(self): + if len(self.datasets) > 0 and self.trainer.dev_data is None: + raise RuntimeError("Trainer has no dev data, you cannot pass extra DataSet to do evaluation.") - :param torch.optim.lr_scheduler._LRScheduler lr_scheduler: PyTorch的lr_scheduler + if len(self.datasets) > 0: + for key, data in self.datasets.items(): + tester = Tester(data=data, model=self.model, + batch_size=self.trainer.kwargs.get('dev_batch_size', self.batch_size), + metrics=self.trainer.metrics, verbose=0, + use_tqdm=self.trainer.test_use_tqdm) + self.testers[key] = tester + + def on_valid_end(self, eval_result, metric_key, optimizer, better_result): + if len(self.testers) > 0: + for key, tester in self.testers.items(): + try: + eval_result = tester.test() + # self.pbar.write("Evaluation on {}:".format(key)) + self.logger.info("Evaluation on {}:".format(key)) + # self.pbar.write(tester._format_eval_results(eval_result)) + self.logger.info(tester._format_eval_results(eval_result)) + except Exception: + # self.pbar.write("Exception happens when evaluate on DataSet named `{}`.".format(key)) + self.logger.info("Exception happens when evaluate on DataSet named `{}`.".format(key)) + + +class LRScheduler(Callback): + """ + 对PyTorch LR Scheduler的包装以使得其可以被Trainer所使用 """ def __init__(self, lr_scheduler): - + """ + :param torch.optim.lr_scheduler._LRScheduler lr_scheduler: PyTorch的lr_scheduler + """ super(LRScheduler, self).__init__() import torch.optim if isinstance(lr_scheduler, torch.optim.lr_scheduler._LRScheduler): @@ -557,13 +690,13 @@ class LRScheduler(Callback): class ControlC(Callback): """ - 别名：:class:`fastNLP.ControlC` :class:`fastNLP.core.callback.ControlC` - - :param bool quit_all: 若为True,则检测到control+C 直接退出程序；否则只退出Trainer + 检测到 control+C 时的反馈 """ def __init__(self, quit_all): - + """ + :param bool quit_all: 若为True,则检测到control+C 直接退出程序；否则只退出Trainer + """ super(ControlC, self).__init__() if type(quit_all) != bool: raise ValueError("In KeyBoardInterrupt, quit_all arguemnt must be a bool.") @@ -586,7 +719,7 @@ class SmoothValue(object): self.smooth = None def add_value(self, val: float) -> None: - "Add `val` to calculate updated smoothed value." + """Add `val` to calculate updated smoothed value.""" self.n += 1 self.mov_avg = self.beta * self.mov_avg + (1 - self.beta) * val self.smooth = self.mov_avg / (1 - self.beta ** self.n) @@ -594,16 +727,15 @@ class SmoothValue(object): class LRFinder(Callback): """ - 别名：:class:`fastNLP.LRFinder` :class:`fastNLP.core.callback.LRFinder` - 用第一个 epoch 找最佳的学习率，从第二个epoch开始应用它 - - :param float start_lr: 学习率下界 - :param float end_lr: 学习率上界 """ def __init__(self, start_lr=1e-6, end_lr=10): + """ + :param float start_lr: 学习率下界 + :param float end_lr: 学习率上界 + """ super(LRFinder, self).__init__() self.start_lr, self.end_lr = start_lr, end_lr @@ -614,8 +746,7 @@ class LRFinder(Callback): self.smooth_value = SmoothValue(0.8) self.opt = None self.find = None - self.loader = ModelLoader() - + @property def lr_gen(self): scale = (self.end_lr - self.start_lr) / self.batch_per_epoch @@ -630,7 +761,7 @@ class LRFinder(Callback): self.opt = self.trainer.optimizer # pytorch optimizer self.opt.param_groups[0]["lr"] = self.start_lr # save model - ModelSaver("tmp").save_pytorch(self.trainer.model, param_only=True) + torch.save(self.model.state_dict(), 'tmp') self.find = True def on_backward_begin(self, loss): @@ -659,14 +790,14 @@ class LRFinder(Callback): self.opt.param_groups[0]["lr"] = self.best_lr self.find = False # reset model - ModelLoader().load_pytorch(self.trainer.model, "tmp") + states = torch.load('tmp') + self.model.load_state_dict(states) + os.remove('tmp') self.pbar.write("Model reset. \nFind best lr={}".format(self.best_lr)) class TensorboardCallback(Callback): """ - 别名：:class:`fastNLP.TensorboardCallback` :class:`fastNLP.core.callback.TensorboardCallback` - 接受以下一个或多个字符串作为参数： - "model" - "loss" @@ -742,13 +873,15 @@ class TensorboardCallback(Callback): class WarmupCallback(Callback): """ 按一定的周期调节Learning rate的大小。 - - :param int,float warmup: 如果warmup为int，则在该step之前，learning rate根据schedule的策略变化; 如果warmup为float， - 如0.1, 则前10%的step是按照schedule策略调整learning rate。 - :param str schedule: 以哪种方式调整。linear: 前warmup的step上升到指定的learning rate(从Trainer中的optimizer处获取的), 后 - warmup的step下降到0； constant前warmup的step上升到指定learning rate，后面的step保持learning rate. """ def __init__(self, warmup=0.1, schedule='constant'): + """ + + :param int,float warmup: 如果warmup为int，则在该step之前，learning rate根据schedule的策略变化; 如果warmup为float， + 如0.1, 则前10%的step是按照schedule策略调整learning rate。 + :param str schedule: 以哪种方式调整。linear: 前warmup的step上升到指定的learning rate(从Trainer中的optimizer处获取的), 后 + warmup的step下降到0； constant前warmup的step上升到指定learning rate，后面的step保持learning rate. + """ super().__init__() self.warmup = max(warmup, 0.) @@ -790,19 +923,23 @@ class WarmupCallback(Callback): class SaveModelCallback(Callback): """ 由于Trainer在训练过程中只会保存最佳的模型， 该callback可实现多种方式的结果存储。 - 会根据训练开始的时间戳在save_dir下建立文件夹，再在文件夹下存放多个模型 - -save_dir - -2019-07-03-15-06-36 - -epoch:0_step:20_{metric_key}:{evaluate_performance}.pt # metric是给定的metric_key, evaluate_performance是性能 - -epoch:1_step:40_{metric_key}:{evaluate_performance}.pt - -2019-07-03-15-10-00 - -epoch:0_step:20_{metric_key}:{evaluate_performance}.pt # metric是给定的metric_key, evaluate_perfomance是性能 - :param str save_dir: 将模型存放在哪个目录下，会在该目录下创建以时间戳命名的目录，并存放模型 - :param int top: 保存dev表现top多少模型。-1为保存所有模型。 - :param bool only_param: 是否只保存模型d饿权重。 - :param save_on_exception: 发生exception时，是否保存一份发生exception的模型。模型名称为epoch:x_step:x_Exception:{exception_name}. + 会根据训练开始的时间戳在save_dir下建立文件夹，再在文件夹下存放多个模型:: + + -save_dir + -2019-07-03-15-06-36 + -epoch:0_step:20_{metric_key}:{evaluate_performance}.pt # metric是给定的metric_key, evaluate_performance是性能 + -epoch:1_step:40_{metric_key}:{evaluate_performance}.pt + -2019-07-03-15-10-00 + -epoch:0_step:20_{metric_key}:{evaluate_performance}.pt # metric是给定的metric_key, evaluate_perfomance是性能 """ def __init__(self, save_dir, top=3, only_param=False, save_on_exception=False): + """ + + :param str save_dir: 将模型存放在哪个目录下，会在该目录下创建以时间戳命名的目录，并存放模型 + :param int top: 保存dev表现top多少模型。-1为保存所有模型。 + :param bool only_param: 是否只保存模型d饿权重。 + :param save_on_exception: 发生exception时，是否保存一份发生exception的模型。模型名称为epoch:x_step:x_Exception:{exception_name}. + """ super().__init__() if not os.path.isdir(save_dir): @@ -850,14 +987,14 @@ class SaveModelCallback(Callback): try: _save_model(self.model, model_name=name, save_dir=self.save_dir, only_param=self.only_param) except Exception as e: - print(f"The following exception:{e} happens when save model to {self.save_dir}.") + logger.error(f"The following exception:{e} happens when save model to {self.save_dir}.") if delete_pair: try: delete_model_path = os.path.join(self.save_dir, delete_pair[1]) if os.path.exists(delete_model_path): os.remove(delete_model_path) except Exception as e: - print(f"Fail to delete model {name} at {self.save_dir} caused by exception:{e}.") + logger.error(f"Fail to delete model {name} at {self.save_dir} caused by exception:{e}.") def on_exception(self, exception): if self.save_on_exception: @@ -868,11 +1005,13 @@ class SaveModelCallback(Callback): class CallbackException(BaseException): """ 当需要通过callback跳出训练的时候可以通过抛出CallbackException并在on_exception中捕获这个值。 - - :param str msg: Exception的信息。 """ def __init__(self, msg): + """ + + :param str msg: Exception的信息。 + """ super(CallbackException, self).__init__(msg) @@ -884,3 +1023,69 @@ class EarlyStopError(CallbackException): def __init__(self, msg): super(EarlyStopError, self).__init__(msg) + + +class EchoCallback(Callback): + def __init__(self, name, out=sys.stdout): + super(EchoCallback, self).__init__() + self.name = name + self.out = out # deprecated + + def __getattribute__(self, item): + if item.startswith('on_'): + logger.info('{}.{} has been called at pid: {}'.format(self.name, item, os.getpid())) + return super(EchoCallback, self).__getattribute__(item) + + +class TesterCallback(Callback): + def __init__(self, data, model, metrics, metric_key=None, batch_size=16, num_workers=None): + super(TesterCallback, self).__init__() + self.tester = Tester(data, model, + metrics=metrics, batch_size=batch_size, + num_workers=num_workers, verbose=0) + # parse metric_key + # increase_better is True. It means the exp result gets better if the indicator increases. + # It is true by default. + self.increase_better = True + if metric_key is not None: + self.increase_better = False if metric_key[0] == "-" else True + self.metric_key = metric_key[1:] if metric_key[0] == "+" or metric_key[0] == "-" else metric_key + else: + self.metric_key = None + self.score = None + + def on_validation(self): + cur_score = self.tester.test() + eval_str = "Evaluation at Epoch {}/{}. Step:{}/{}. - {}".format( + self.epoch, self.n_epochs, self.step, self.n_steps, + self.tester._format_eval_results(cur_score)) + self.logger.info(eval_str) + is_better = self.compare_better(cur_score) + if is_better: + self.score = cur_score + return cur_score, is_better + + def _get_score(self, metric_dict, key): + for metric in metric_dict.items(): + if key in metric: + return metric[key] + return None + + def compare_better(self, a): + if self.score is None: + return True + if self.metric_key is None: + self.metric_key = list(list(self.score.values())[0].keys())[0] + k = self.metric_key + score = self._get_score(self.score, k) + new_score = self._get_score(a, k) + if score is None or new_score is None: + return False + if self.increase_better: + return score <= new_score + else: + return score >= new_score + + def on_train_end(self): + self.logger.info('Evaluate on training ends.') + self.on_validation() diff --git a/fastNLP/core/const.py b/fastNLP/core/const.py index 89ff51a2..ad5d1f1e 100644 --- a/fastNLP/core/const.py +++ b/fastNLP/core/const.py @@ -1,3 +1,13 @@ +""" +.. todo:: + doc +""" + +__all__ = [ + "Const" +] + + class Const: """ fastNLP中field命名常量。 @@ -7,12 +17,14 @@ class Const: 具体列表:: - INPUT 模型的序列输入 words（复数words1, words2） - CHAR_INPUT 模型character输入 chars（复数chars1， chars2） - INPUT_LEN 序列长度 seq_len（复数seq_len1，seq_len2） - OUTPUT 模型输出 pred（复数pred1， pred2） - TARGET 真实目标 target（复数target1，target2） - LOSS 损失函数 loss （复数loss1，loss2） + INPUT 模型的序列输入 words（具有多列words时，依次使用words1, words2, ） + CHAR_INPUT 模型character输入 chars（具有多列chars时，依次使用chars1， chars2） + INPUT_LEN 序列长度 seq_len（具有多列seq_len时，依次使用seq_len1，seq_len2） + OUTPUT 模型输出 pred（具有多列pred时，依次使用pred1， pred2） + TARGET 真实目标 target（具有多列target时，依次使用target1，target2） + LOSS 损失函数 loss （具有多列loss时，依次使用loss1，loss2） + RAW_WORD 原文的词 raw_words (具有多列raw_words时，依次使用raw_words1, raw_words2) + RAW_CHAR 原文的字 raw_chars (具有多列raw_chars时，依次使用raw_chars1, raw_chars2) """ INPUT = 'words' @@ -21,37 +33,49 @@ class Const: OUTPUT = 'pred' TARGET = 'target' LOSS = 'loss' - + RAW_WORD = 'raw_words' + RAW_CHAR = 'raw_chars' + @staticmethod def INPUTS(i): """得到第 i 个 ``INPUT`` 的命名""" i = int(i) + 1 return Const.INPUT + str(i) - + @staticmethod def CHAR_INPUTS(i): """得到第 i 个 ``CHAR_INPUT`` 的命名""" i = int(i) + 1 return Const.CHAR_INPUT + str(i) - + + @staticmethod + def RAW_WORDS(i): + i = int(i) + 1 + return Const.RAW_WORD + str(i) + + @staticmethod + def RAW_CHARS(i): + i = int(i) + 1 + return Const.RAW_CHAR + str(i) + @staticmethod def INPUT_LENS(i): """得到第 i 个 ``INPUT_LEN`` 的命名""" i = int(i) + 1 return Const.INPUT_LEN + str(i) - + @staticmethod def OUTPUTS(i): """得到第 i 个 ``OUTPUT`` 的命名""" i = int(i) + 1 return Const.OUTPUT + str(i) - + @staticmethod def TARGETS(i): """得到第 i 个 ``TARGET`` 的命名""" i = int(i) + 1 return Const.TARGET + str(i) - + @staticmethod def LOSSES(i): """得到第 i 个 ``LOSS`` 的命名""" diff --git a/fastNLP/core/dataset.py b/fastNLP/core/dataset.py index 7b7fa87a..2b548f22 100644 --- a/fastNLP/core/dataset.py +++ b/fastNLP/core/dataset.py @@ -288,29 +288,33 @@ __all__ = [ ] import _pickle as pickle -import warnings +from copy import deepcopy import numpy as np +from ._logger import logger +from .const import Const +from .field import AppendToTargetOrInputException from .field import AutoPadder from .field import FieldArray +from .field import SetInputOrTargetException from .instance import Instance from .utils import _get_func_signature -from .field import AppendToTargetOrInputException -from .field import SetInputOrTargetException +from .utils import pretty_table_printer +from prettytable import PrettyTable + class DataSet(object): """ - 别名：:class:`fastNLP.DataSet` :class:`fastNLP.core.dataset.DataSet` - fastNLP的数据容器，详细的使用方法见文档 :doc:`fastNLP.core.dataset` - - :param data: 如果为dict类型，则每个key的value应该为等长的list; 如果为list， - 每个元素应该为具有相同field的 :class:`~fastNLP.Instance` 。 - """ - + def __init__(self, data=None): + """ + + :param data: 如果为dict类型，则每个key的value应该为等长的list; 如果为list， + 每个元素应该为具有相同field的 :class:`~fastNLP.Instance` 。 + """ self.field_arrays = {} if data is not None: if isinstance(data, dict): @@ -324,41 +328,45 @@ class DataSet(object): for ins in data: assert isinstance(ins, Instance), "Must be Instance type, not {}.".format(type(ins)) self.append(ins) - + else: raise ValueError("data only be dict or list type.") - + def __contains__(self, item): return item in self.field_arrays - + def __iter__(self): def iter_func(): for idx in range(len(self)): yield self[idx] - + return iter_func() - + def _inner_iter(self): class Iter_ptr: def __init__(self, dataset, idx): self.dataset = dataset self.idx = idx - + def __getitem__(self, item): assert item in self.dataset.field_arrays, "no such field:{} in Instance {}".format(item, self.dataset[ self.idx]) assert self.idx < len(self.dataset.field_arrays[item]), "index:{} out of range".format(self.idx) return self.dataset.field_arrays[item][self.idx] - + + def items(self): + ins = self.dataset[self.idx] + return ins.items() + def __repr__(self): return self.dataset[self.idx].__repr__() - + def inner_iter_func(): for idx in range(len(self)): yield Iter_ptr(self, idx) - + return inner_iter_func() - + def __getitem__(self, idx): """给定int的index，返回一个Instance; 给定slice，返回包含这个slice内容的新的DataSet。 @@ -391,20 +399,20 @@ class DataSet(object): return dataset else: raise KeyError("Unrecognized type {} for idx in __getitem__ method".format(type(idx))) - + def __getattr__(self, item): # Not tested. Don't use !! if item == "field_arrays": raise AttributeError if isinstance(item, str) and item in self.field_arrays: return self.field_arrays[item] - + def __setstate__(self, state): self.__dict__ = state - + def __getstate__(self): return self.__dict__ - + def __len__(self): """Fetch the length of the dataset. @@ -414,16 +422,66 @@ class DataSet(object): return 0 field = iter(self.field_arrays.values()).__next__() return len(field) - - def __inner_repr__(self): - if len(self) < 20: - return ",\n".join([ins.__repr__() for ins in self]) - else: - return self[:5].__inner_repr__() + "\n...\n" + self[-5:].__inner_repr__() - + def __repr__(self): - return "DataSet(" + self.__inner_repr__() + ")" - + return str(pretty_table_printer(self)) + + def print_field_meta(self): + """ + 输出当前field的meta信息, 形似下列的输出 + + +-------------+-------+-------+ + | field_names | x | y | + +-------------+-------+-------+ + | is_input | True | False | + | is_target | False | False | + | ignore_type | False | | + | pad_value | 0 | | + +-------------+-------+-------+ + + field_names: DataSet中field的名称 + is_input: field是否为input + is_target: field是否为target + ignore_type: 是否忽略该field的type, 一般仅在该field至少为input或target时才有意义 + pad_value: 该field的pad的值，仅在该field为input或target时有意义 + + :return: + """ + if len(self.field_arrays)>0: + field_names = ['field_names'] + is_inputs = ['is_input'] + is_targets = ['is_target'] + pad_values = ['pad_value'] + ignore_types = ['ignore_type'] + + for name, field_array in self.field_arrays.items(): + field_names.append(name) + if field_array.is_input: + is_inputs.append(True) + else: + is_inputs.append(False) + if field_array.is_target: + is_targets.append(True) + else: + is_targets.append(False) + + if (field_array.is_input or field_array.is_target) and field_array.padder is not None: + pad_values.append(field_array.padder.get_pad_val()) + else: + pad_values.append(' ') + + if field_array._ignore_type: + ignore_types.append(True) + elif field_array.is_input or field_array.is_target: + ignore_types.append(False) + else: + ignore_types.append(' ') + table = PrettyTable(field_names=field_names) + fields = [is_inputs, is_targets, ignore_types, pad_values] + for field in fields: + table.add_row(field) + logger.info(table) + def append(self, instance): """ 将一个instance对象append到DataSet后面。 @@ -446,9 +504,9 @@ class DataSet(object): try: self.field_arrays[name].append(field) except AppendToTargetOrInputException as e: - print(f"Cannot append to field:{name}.") + logger.error(f"Cannot append to field:{name}.") raise e - + def add_fieldarray(self, field_name, fieldarray): """ 将fieldarray添加到DataSet中. @@ -463,7 +521,7 @@ class DataSet(object): raise RuntimeError(f"The field to add must have the same size as dataset. " f"Dataset size {len(self)} != field size {len(fieldarray)}") self.field_arrays[field_name] = fieldarray - + def add_field(self, field_name, fields, padder=AutoPadder(), is_input=False, is_target=False, ignore_type=False): """ 新增一个field @@ -475,19 +533,19 @@ class DataSet(object): :param bool is_target: 新加入的field是否是target :param bool ignore_type: 是否忽略对新加入的field的类型检查 """ - + if len(self.field_arrays) != 0: if len(self) != len(fields): raise RuntimeError(f"The field to add must have the same size as dataset. " f"Dataset size {len(self)} != field size {len(fields)}") self.field_arrays[field_name] = FieldArray(field_name, fields, is_target=is_target, is_input=is_input, padder=padder, ignore_type=ignore_type) - + def delete_instance(self, index): """ 删除第index个instance - :param int index: 需要删除的instance的index，从0开始 + :param int index: 需要删除的instance的index，序号从0开始。 """ assert isinstance(index, int), "Only integer supported." if len(self) <= index: @@ -497,7 +555,8 @@ class DataSet(object): else: for field in self.field_arrays.values(): field.pop(index) - + return self + def delete_field(self, field_name): """ 删除名为field_name的field @@ -505,7 +564,22 @@ class DataSet(object): :param str field_name: 需要删除的field的名称. """ self.field_arrays.pop(field_name) - + return self + + def copy_field(self, field_name, new_field_name): + """ + 深度copy名为field_name的field到new_field_name + + :param str field_name: 需要copy的field。 + :param str new_field_name: copy生成的field名称 + :return: self + """ + if not self.has_field(field_name): + raise KeyError(f"Field:{field_name} not found in DataSet.") + fieldarray = deepcopy(self.get_field(field_name)) + self.add_fieldarray(field_name=new_field_name, fieldarray=fieldarray) + return self + def has_field(self, field_name): """ 判断DataSet中是否有名为field_name这个field @@ -516,7 +590,7 @@ class DataSet(object): if isinstance(field_name, str): return field_name in self.field_arrays return False - + def get_field(self, field_name): """ 获取field_name这个field @@ -527,7 +601,7 @@ class DataSet(object): if field_name not in self.field_arrays: raise KeyError("Field name {} not found in DataSet".format(field_name)) return self.field_arrays[field_name] - + def get_all_fields(self): """ 返回一个dict，key为field_name, value为对应的 :class:`~fastNLP.FieldArray` @@ -535,7 +609,7 @@ class DataSet(object): :return dict: 返回如上所述的字典 """ return self.field_arrays - + def get_field_names(self) -> list: """ 返回一个list，包含所有 field 的名字 @@ -543,7 +617,7 @@ class DataSet(object): :return list: 返回如上所述的列表 """ return sorted(self.field_arrays.keys()) - + def get_length(self): """ 获取DataSet的元素数量 @@ -551,22 +625,22 @@ class DataSet(object): :return: int: DataSet中Instance的个数。 """ return len(self) - - def rename_field(self, old_name, new_name): + + def rename_field(self, field_name, new_field_name): """ 将某个field重新命名. - :param str old_name: 原来的field名称。 - :param str new_name: 修改为new_name。 + :param str field_name: 原来的field名称。 + :param str new_field_name: 修改为new_name。 """ - if old_name in self.field_arrays: - self.field_arrays[new_name] = self.field_arrays.pop(old_name) - self.field_arrays[new_name].name = new_name + if field_name in self.field_arrays: + self.field_arrays[new_field_name] = self.field_arrays.pop(field_name) + self.field_arrays[new_field_name].name = new_field_name else: - raise KeyError("DataSet has no field named {}.".format(old_name)) + raise KeyError("DataSet has no field named {}.".format(field_name)) return self - - def set_target(self, *field_names, flag=True): + + def set_target(self, *field_names, flag=True, use_1st_ins_infer_dim_type=True): """ 将field_names的field设置为target @@ -577,19 +651,23 @@ class DataSet(object): :param str field_names: field的名称 :param bool flag: 将field_name的target状态设置为flag + :param bool use_1st_ins_infer_dim_type: 如果为True，将不会check该列是否所有数据都是同样的维度，同样的类型。将直接使用第一 + 行的数据进行类型和维度推断本列的数据的类型和维度。 """ assert isinstance(flag, bool), "Only bool type supported." for name in field_names: if name in self.field_arrays: try: + self.field_arrays[name]._use_1st_ins_infer_dim_type = bool(use_1st_ins_infer_dim_type) self.field_arrays[name].is_target = flag except SetInputOrTargetException as e: - print(f"Cannot set field:{name} as target.") + logger.error(f"Cannot set field:{name} as target.") raise e else: raise KeyError("{} is not a valid field name.".format(name)) - - def set_input(self, *field_names, flag=True): + return self + + def set_input(self, *field_names, flag=True, use_1st_ins_infer_dim_type=True): """ 将field_names的field设置为input:: @@ -598,17 +676,21 @@ class DataSet(object): :param str field_names: field的名称 :param bool flag: 将field_name的input状态设置为flag + :param bool use_1st_ins_infer_dim_type: 如果为True，将不会check该列是否所有数据都是同样的维度，同样的类型。将直接使用第一 + 行的数据进行类型和维度推断本列的数据的类型和维度。 """ for name in field_names: if name in self.field_arrays: try: + self.field_arrays[name]._use_1st_ins_infer_dim_type = bool(use_1st_ins_infer_dim_type) self.field_arrays[name].is_input = flag except SetInputOrTargetException as e: - print(f"Cannot set field:{name} as input, exception happens at the {e.index} value.") + logger.error(f"Cannot set field:{name} as input, exception happens at the {e.index} value.") raise e else: raise KeyError("{} is not a valid field name.".format(name)) - + return self + def set_ignore_type(self, *field_names, flag=True): """ 将field设置为忽略类型状态。当某个field被设置了ignore_type, 则在被设置为target或者input时将不进行类型检查， @@ -624,7 +706,8 @@ class DataSet(object): self.field_arrays[name].ignore_type = flag else: raise KeyError("{} is not a valid field name.".format(name)) - + return self + def set_padder(self, field_name, padder): """ 为field_name设置padder:: @@ -639,7 +722,8 @@ class DataSet(object): if field_name not in self.field_arrays: raise KeyError("There is no field named {}.".format(field_name)) self.field_arrays[field_name].set_padder(padder) - + return self + def set_pad_val(self, field_name, pad_val): """ 为某个field设置对应的pad_val. @@ -650,7 +734,8 @@ class DataSet(object): if field_name not in self.field_arrays: raise KeyError("There is no field named {}.".format(field_name)) self.field_arrays[field_name].set_pad_val(pad_val) - + return self + def get_input_name(self): """ 返回所有is_input被设置为True的field名称 @@ -658,7 +743,7 @@ class DataSet(object): :return list: 里面的元素为被设置为input的field名称 """ return [name for name, field in self.field_arrays.items() if field.is_input] - + def get_target_name(self): """ 返回所有is_target被设置为True的field名称 @@ -666,7 +751,7 @@ class DataSet(object): :return list: 里面的元素为被设置为target的field名称 """ return [name for name, field in self.field_arrays.items() if field.is_target] - + def apply_field(self, func, field_name, new_field_name=None, **kwargs): """ 将DataSet中的每个instance中的名为 `field_name` 的field传给func，并获取它的返回值。 @@ -695,16 +780,16 @@ class DataSet(object): results.append(func(ins[field_name])) except Exception as e: if idx != -1: - print("Exception happens at the `{}`th instance.".format(idx)) + logger.error("Exception happens at the `{}`th(from 1) instance.".format(idx + 1)) raise e if not (new_field_name is None) and len(list(filter(lambda x: x is not None, results))) == 0: # all None raise ValueError("{} always return None.".format(_get_func_signature(func=func))) - + if new_field_name is not None: self._add_apply_field(results, new_field_name, kwargs) - + return results - + def _add_apply_field(self, results, new_field_name, kwargs): """ 将results作为加入到新的field中，field名称为new_field_name @@ -736,7 +821,7 @@ class DataSet(object): self.add_field(field_name=new_field_name, fields=results, is_input=extra_param.get("is_input", None), is_target=extra_param.get("is_target", None), ignore_type=extra_param.get("ignore_type", False)) - + def apply(self, func, new_field_name=None, **kwargs): """ 将DataSet中每个instance传入到func中，并获取它的返回值. @@ -760,20 +845,21 @@ class DataSet(object): results = [] for idx, ins in enumerate(self._inner_iter()): results.append(func(ins)) - except Exception as e: + except BaseException as e: if idx != -1: - print("Exception happens at the `{}`th instance.".format(idx)) + logger.error("Exception happens at the `{}`th instance.".format(idx)) raise e + # results = [func(ins) for ins in self._inner_iter()] if not (new_field_name is None) and len(list(filter(lambda x: x is not None, results))) == 0: # all None raise ValueError("{} always return None.".format(_get_func_signature(func=func))) - + if new_field_name is not None: self._add_apply_field(results, new_field_name, kwargs) - + return results - def add_seq_len(self, field_name:str, new_field_name='seq_len'): + def add_seq_len(self, field_name: str, new_field_name=Const.INPUT_LEN): """ 将使用len()直接对field_name中每个元素作用，将其结果作为seqence length, 并放入seq_len这个field。 @@ -810,7 +896,7 @@ class DataSet(object): return dataset else: return DataSet() - + def split(self, ratio, shuffle=True): """ 将DataSet按照ratio的比例拆分，返回两个DataSet @@ -836,51 +922,9 @@ class DataSet(object): for field_name in self.field_arrays: train_set.field_arrays[field_name].to(self.field_arrays[field_name]) dev_set.field_arrays[field_name].to(self.field_arrays[field_name]) - + return train_set, dev_set - - @classmethod - def read_csv(cls, csv_path, headers=None, sep=",", dropna=True): - r""" - .. warning:: - 此方法会在下个版本移除，请使用 :class:`fastNLP.io.CSVLoader` - - 从csv_path路径下以csv的格式读取数据。 - :param str csv_path: 从哪里读取csv文件 - :param list[str] headers: 如果为None，则使用csv文件的第一行作为header; 如果传入list(str), 则元素的个数必须 - 与csv文件中每行的元素个数相同。 - :param str sep: 分割符 - :param bool dropna: 是否忽略与header数量不一致行。 - :return: 读取后的 :class:`~fastNLP.读取后的DataSet`。 - """ - warnings.warn('DataSet.read_csv is deprecated, use CSVLoader instead', - category=DeprecationWarning) - with open(csv_path, "r", encoding='utf-8') as f: - start_idx = 0 - if headers is None: - headers = f.readline().rstrip('\r\n') - headers = headers.split(sep) - start_idx += 1 - else: - assert isinstance(headers, (list, tuple)), "headers should be list or tuple, not {}.".format( - type(headers)) - _dict = {} - for col in headers: - _dict[col] = [] - for line_idx, line in enumerate(f, start_idx): - contents = line.rstrip('\r\n').split(sep) - if len(contents) != len(headers): - if dropna: - continue - else: - # TODO change error type - raise ValueError("Line {} has {} parts, while header has {} parts." \ - .format(line_idx, len(contents), len(headers))) - for header, content in zip(headers, contents): - _dict[header].append(content) - return cls(_dict) - def save(self, path): """ 保存DataSet. @@ -889,7 +933,7 @@ class DataSet(object): """ with open(path, 'wb') as f: pickle.dump(self, f) - + @staticmethod def load(path): r""" diff --git a/fastNLP/core/dist_trainer.py b/fastNLP/core/dist_trainer.py new file mode 100644 index 00000000..3a293447 --- /dev/null +++ b/fastNLP/core/dist_trainer.py @@ -0,0 +1,356 @@ +"""undocumented +正在开发中的分布式训练代码 +""" +import logging +import os +import time +from datetime import datetime + +import torch +import torch.cuda +import torch.distributed as dist +import torch.optim +from pkg_resources import parse_version +from torch.nn.parallel import DistributedDataParallel as DDP +from torch.utils.data.distributed import DistributedSampler +from tqdm import tqdm + +from ._logger import logger +from .batch import DataSetIter, BatchIter +from .callback import DistCallbackManager, CallbackException, TesterCallback +from .dataset import DataSet +from .losses import _prepare_losser +from .optimizer import Optimizer +from .utils import _build_args +from .utils import _get_func_signature +from .utils import _move_dict_value_to_device + +__all__ = [ + 'get_local_rank', + 'DistTrainer', +] + + +def get_local_rank(): + if 'LOCAL_RANK' in os.environ: + return int(os.environ['LOCAL_RANK']) + from argparse import ArgumentParser + parser = ArgumentParser() + parser.add_argument('--local_rank', type=int) + args, _ = parser.parse_known_args() + if 'local_rank' in args and args.local_rank: + os.environ['LOCAL_RANK'] = str(args.local_rank) # for multiple calls for this function + return args.local_rank + raise RuntimeError('Please use "python -m torch.distributed.launch --nproc_per_node=N train_script.py') + + +class DistTrainer(): + """ + Distributed Trainer that support distributed and mixed precision training + """ + def __init__(self, train_data, model, optimizer=None, loss=None, + callbacks_all=None, callbacks_master=None, + batch_size_per_gpu=8, n_epochs=1, + num_workers=1, drop_last=False, + dev_data=None, metrics=None, metric_key=None, + update_every=1, print_every=10, validate_every=-1, + save_every=-1, save_path=None, device='auto', + fp16='', backend=None, init_method=None): + + assert device in ['auto', 'cuda', 'cpu'], "Please set correct device in [auto', 'cuda', 'cpu']" + if device == 'auto': + device = 'cuda' if torch.cuda.is_available() else 'cpu' + if backend is None: + backend = 'nccl' if device == 'cuda' else 'gloo' + + # init distributed + if device == 'cuda': + torch.cuda.set_device(get_local_rank()) + self.device = torch.device("cuda", get_local_rank()) + else: + self.device = torch.device(device) + + dist.init_process_group(backend=backend, init_method=init_method) + self.world_size = dist.get_world_size() + self.rank = dist.get_rank() # unique id for each process + + self.model = model + self.train_data = train_data + self.batch_size_per_gpu = int(batch_size_per_gpu) + self.n_epochs = int(n_epochs) + self.num_data_workers = int(num_workers) + self.drop_last = drop_last + self.update_every = int(update_every) + self.print_every = int(print_every) + self.validate_every = int(validate_every) + self.save_every = int(save_every) + self.save_path = save_path + self.losser = _prepare_losser(loss) + self.fp16 = fp16 + self.init_method = init_method + self.backend = backend + self.local_rank = get_local_rank() + self._forward_func = model.forward + self.callback_manager = DistCallbackManager( + env={"trainer": self}, callbacks_all=callbacks_all, + callbacks_master=callbacks_master) + self.metric_key = metric_key + + model.to(self.device) + optimizer = self._get_optimizer(optimizer) + + # init fp16, must before DataParallel init + if len(self.fp16): + assert isinstance(self.fp16, str), "Please set Apex AMP optimization level selected in ['O0', 'O1', 'O2', 'O3']" + try: + from apex import amp + except ImportError: + raise ImportError("Please install apex from https://www.github.com/nvidia/apex to use fp16 training.") + assert torch.backends.cudnn.enabled, "Amp requires cudnn backend to be enabled." + assert device == 'cuda', "Amp requires cuda device" + model, optimizer = amp.initialize(model, optimizer, opt_level=self.fp16) + + # init DataParallel + if parse_version(torch.__version__)>=parse_version('1.1'): + self.model = DDP(model, device_ids=[self.local_rank], + output_device=self.local_rank, find_unused_parameters=True) + else: + self.model = DDP(model, device_ids=[self.local_rank], + output_device=self.local_rank) + + self.optimizer = optimizer + self.sampler = DistributedSampler(self.train_data) + self.data_iterator = self._get_data_iter(self.train_data) + self.n_steps = self._get_n_steps() + + # for evaluation, only run eval on master proc + if dev_data and metrics: + cb = TesterCallback( + dev_data, model, metrics, + batch_size=batch_size_per_gpu, num_workers=num_workers) + self.callback_manager.add_callback([cb], master=True) + + # Setup logging + dist.barrier() + self.start_time = datetime.now().strftime('%m_%d_%Y-%H_%M') + if self.save_path: + self.cp_save_path = os.path.join(self.save_path, 'checkpoints', self.start_time) + else: + self.cp_save_path = None + + # use INFO in the master, WARN for others + logger.setLevel(logging.INFO if self.is_master else logging.WARNING) + self.logger = logger + self.logger.info("Setup Distributed Trainer") + self.logger.warning("Process pid: {}, rank: {}, local rank: {}, device: {}, fp16: {}".format( + os.getpid(), self.rank, self.local_rank, self.device, self.fp16 if self.fp16 else False)) + self.logger.info("Num of processes: {}".format(self.world_size)) + self.logger.info("Use device: {}".format(device)) + self.logger.info("Training with fp16: {}, optimization level: {}".format( + len(self.fp16) > 0, self.fp16 if self.fp16 else None)) + + def _get_n_steps(self): + batch_size = self.world_size * self.batch_size_per_gpu + return (len(self.train_data) // batch_size + int( + len(self.train_data) % batch_size != 0)) * int(self.drop_last == 0) * self.n_epochs + + def _get_data_iter(self, dataset): + if isinstance(dataset, DataSet): + return DataSetIter( + dataset=dataset, batch_size=self.batch_size_per_gpu, + num_workers=self.num_data_workers, sampler=self.sampler, + drop_last=self.drop_last + ) + elif isinstance(dataset, BatchIter): + return dataset + else: + raise TypeError("train_data type {} not support".format(type(dataset))) + + def _get_optimizer(self, optimizer): + if isinstance(optimizer, torch.optim.Optimizer): + return optimizer + elif isinstance(optimizer, Optimizer): + return optimizer.construct_from_pytorch(self.model.parameters()) + elif optimizer is None: + return torch.optim.Adam(self.model.parameters(), lr=4e-3) + else: + raise TypeError("optimizer can only be torch.optim.Optimizer type, not {}.".format(type(optimizer))) + + @property + def is_master(self): + return self.rank == 0 + + def train(self, on_exception='auto'): + try: + self.logger.info("###### Training epochs started ######") + self.logger.info('Total epochs: %d'% self.n_epochs) + self.logger.info('Total steps: %d'% self.n_steps) + self.logger.info('Num instances per GPU %d'% self.batch_size_per_gpu) + self.logger.info('Total batch_size: %d'% self.batch_size_per_gpu * dist.get_world_size()) + self.logger.info('Total num of samples: %d'% len(self.train_data)) + self.logger.info("Num of callbacks for all workers: {}".format( + len(self.callback_manager.callbacks_all))) + self.logger.info("Num of callbacks for master workers: {}".format( + len(self.callback_manager.callbacks_master))) + self.logger.info("Callbacks for all workers: {}".format( + [repr(cb) for cb in self.callback_manager.callbacks_all])) + self.logger.info("Callbacks for master workers: {}".format( + [repr(cb) for cb in self.callback_manager.callbacks_master])) + + start_time = time.time() + results = {} + if self.n_epochs <= 0: + self.logger.info("Training epoch is {}, nothing was done.".format(self.n_epochs)) + results['seconds'] = 0. + return results + + try: + self.callback_manager.on_train_begin() + self._train() + self.callback_manager.on_train_end() + + except BaseException as e: + self.callback_manager.on_exception(e) + if on_exception == 'auto': + if not isinstance(e, (CallbackException, KeyboardInterrupt)): + raise e + else: + self.logger.info('Catch {}, ignored.'.format(e.__class__.__name__)) + elif on_exception == 'raise': + raise e + + results['seconds'] = round(time.time() - start_time, 2) + self.logger.info("###### Train finished ######") + self.logger.info('Total train time: {} seconds.'. format(results['seconds'])) + return results + finally: + self.close() + + def _train(self): + if self.fp16: + # skip check, done in __init__() + from apex import amp + self.step = 0 + self.epoch = 0 + self.pbar = tqdm(total=self.n_steps, postfix='loss:{0:<6.5f}', + leave=False, dynamic_ncols=True, disable=not self.is_master) + pbar = self.pbar + avg_loss = 0 + data_iterator = self.data_iterator + self.model.zero_grad() + for epoch in range(1, self.n_epochs + 1): + self.epoch = epoch + pbar.set_description_str(desc="Epoch {}/{}".format(epoch, self.n_epochs)) + # early stopping + self.callback_manager.on_epoch_begin() + for batch_x, batch_y in data_iterator: + self.model.train() + self.step += 1 + _move_dict_value_to_device(batch_x, batch_y, device=self.device) + indices = data_iterator.get_batch_indices() + # negative sampling; replace unknown; re-weight batch_y + self.callback_manager.on_batch_begin(batch_x, batch_y, indices) + prediction = self._data_forward(self.model, batch_x) + + # edit prediction + self.callback_manager.on_loss_begin(batch_y, prediction) + loss = self._compute_loss(prediction, batch_y) + avg_loss += loss.item() + + # Is loss NaN or inf? requires_grad = False + self.callback_manager.on_backward_begin(loss) + + if self.fp16: + with amp.scale_loss(loss, self.optimizer) as scale_loss: + scale_loss.backward() + else: + loss.backward() + + self.callback_manager.on_backward_end() + + self._update() + self.callback_manager.on_step_end() + + if self.step % self.print_every == 0: + avg_loss = float(avg_loss) / self.print_every + print_output = "loss:{:<6.5f}".format(avg_loss) + pbar.update(self.print_every) + pbar.set_postfix_str(print_output) + avg_loss = 0 + + self.callback_manager.on_batch_end() + + if (self.validate_every > 0 and self.step % self.validate_every == 0): + self._do_validation() + + if self.cp_save_path and \ + self.save_every > 0 and \ + self.step % self.save_every == 0: + self.save_check_point() + + # ================= mini-batch end ==================== # + if self.validate_every < 0: + self._do_validation() + + if self.save_every < 0 and self.cp_save_path: + self.save_check_point() + # lr decay; early stopping + self.callback_manager.on_epoch_end() + # =============== epochs end =================== # + pbar.close() + self.pbar = None + # ============ tqdm end ============== # + + def _update(self): + """Perform weight update on a model. + + """ + if self.step % self.update_every == 0: + self.optimizer.step() + self.model.zero_grad() + + def _data_forward(self, network, x): + x = _build_args(self._forward_func, **x) + y = network(**x) + if not isinstance(y, dict): + raise TypeError( + f"The return value of {_get_func_signature(self._forward_func)} should be dict, got {type(y)}.") + return y + + def _compute_loss(self, predict, truth): + """Compute loss given prediction and ground truth. + + :param predict: prediction dict, produced by model.forward + :param truth: ground truth dict, produced by batch_y + :return: a scalar + """ + loss = self.losser(predict, truth) + if self.update_every > 1: + loss = loss / self.update_every + return loss.mean() + + def save_check_point(self, only_params=False): + # only master save models + if self.is_master: + os.makedirs(self.cp_save_path, exist_ok=True) + path = os.path.join(self.cp_save_path, 'checkpoint-{}.bin'.format(self.step)) + self.logger.info("Save checkpoint to {}".format(path)) + model_to_save = self.model.module + if only_params: + model_to_save = model_to_save.state_dict() + torch.save(model_to_save, path) + + def _do_validation(self): + self.callback_manager.on_valid_begin() + eval_res = self.callback_manager.on_validation() + eval_res = list(filter(lambda x: x is not None, eval_res)) + if len(eval_res): + eval_res, is_better = list(zip(*eval_res)) + else: + eval_res, is_better = None, None + self.callback_manager.on_valid_end( + eval_res, self.metric_key, self.optimizer, is_better) + dist.barrier() + + def close(self): + dist.destroy_process_group() diff --git a/fastNLP/core/field.py b/fastNLP/core/field.py index bba854f5..1835bafa 100644 --- a/fastNLP/core/field.py +++ b/fastNLP/core/field.py @@ -1,73 +1,91 @@ +""" +.. todo:: + doc +""" +__all__ = [ + "Padder", + "AutoPadder", + "EngChar2DPadder", +] -from numbers import Number -import torch -import numpy as np -from typing import Any from abc import abstractmethod -from copy import deepcopy from collections import Counter +from copy import deepcopy +from numbers import Number +from typing import Any + +import numpy as np +import torch + +from ._logger import logger +from .utils import _is_iterable + class SetInputOrTargetException(Exception): def __init__(self, msg, index=None, field_name=None): super().__init__(msg) self.msg = msg self.index = index # 标示在哪个数据遭遇到问题了 - self.field_name = field_name # 标示当前field的名称 + self.field_name = field_name # 标示当前field的名称 + class AppendToTargetOrInputException(Exception): def __init__(self, msg, index=None, field_name=None): super().__init__(msg) self.msg = msg self.index = index # 标示在哪个数据遭遇到问题了 - self.field_name = field_name # 标示当前field的名称 + self.field_name = field_name # 标示当前field的名称 + class FieldArray: - def __init__(self, name, content, is_target=False, is_input=False, padder=None, ignore_type=False): - if len(content)==0: + def __init__(self, name, content, is_target=False, is_input=False, padder=None, ignore_type=False, + use_1st_ins_infer_dim_type=True): + if len(content) == 0: raise RuntimeError("Empty fieldarray is not allowed.") _content = content try: _content = list(_content) except BaseException as e: - print(f"Cannot convert content(of type:{type(content)}) into list.") + logger.error(f"Cannot convert content(of type:{type(content)}) into list.") raise e self.name = name self.content = _content self._ignore_type = ignore_type # 根据input的情况设置input，target等 - self._cell_ndim = None # 多少维度 + self._cell_ndim = None # 多少维度， 如果value是1, dim为0; 如果value是[1, 2], dim=2 self.dtype = None # 最内层的element都是什么类型的 + self._use_1st_ins_infer_dim_type = bool(use_1st_ins_infer_dim_type) self._is_input = False self._is_target = False - + if is_input: self.is_input = is_input if is_target: self.is_target = is_target - + if padder is None: padder = AutoPadder(pad_val=0) else: assert isinstance(padder, Padder), "padder must be of type fastNLP.Padder." padder = deepcopy(padder) self.set_padder(padder) - + @property def ignore_type(self): return self._ignore_type - + @ignore_type.setter def ignore_type(self, value): if value: self._cell_ndim = None self.dtype = None self._ignore_type = value - + @property def is_input(self): return self._is_input - + @is_input.setter def is_input(self, value): """ @@ -77,16 +95,16 @@ class FieldArray: if value is True and \ self._is_target is False and \ self._ignore_type is False: - self._check_dtype_and_ndim() + self._check_dtype_and_ndim(only_check_1st_ins_dim_type=self._use_1st_ins_infer_dim_type) if value is False and self._is_target is False: self.dtype = None self._cell_ndim = None self._is_input = value - + @property def is_target(self): return self._is_target - + @is_target.setter def is_target(self, value): """ @@ -95,70 +113,82 @@ class FieldArray: if value is True and \ self._is_input is False and \ self._ignore_type is False: - self._check_dtype_and_ndim() + self._check_dtype_and_ndim(only_check_1st_ins_dim_type=self._use_1st_ins_infer_dim_type) if value is False and self._is_input is False: self.dtype = None self._cell_ndim = None self._is_target = value - - def _check_dtype_and_ndim(self): + + def _check_dtype_and_ndim(self, only_check_1st_ins_dim_type=True): """ 检查当前content所有的element是否是同一个类型，且是否每个元素具有相同的维度。通过的话，设置_cell_ndim与_ele_type属性；没有 通过将直接报错. + :param bool only_check_1st_ins_dim_type: 是否只检查第一个元素的type和dim :return: """ cell_0 = self.content[0] index = 0 try: type_0, dim_0 = _get_ele_type_and_dim(cell_0) - for cell in self.content[1:]: - index += 1 - type_i, dim_i = _get_ele_type_and_dim(cell) - if type_i!=type_0: - raise SetInputOrTargetException("Type:{} in index {} is different from the first element with type:{}." - ".".format(type_i, index, type_0)) - if dim_0!=dim_i: - raise SetInputOrTargetException("Dimension:{} in index {} is different from the first element with " - "dimension:{}.".format(dim_i, index, dim_0)) + if not only_check_1st_ins_dim_type: + for cell in self.content[1:]: + index += 1 + type_i, dim_i = _get_ele_type_and_dim(cell) + if type_i != type_0: + raise SetInputOrTargetException( + "Type:{} in index {} is different from the first element with type:{}." + ".".format(type_i, index, type_0)) + if dim_0 != dim_i: + raise SetInputOrTargetException( + "Dimension:{} in index {} is different from the first element with " + "dimension:{}.".format(dim_i, index, dim_0)) self._cell_ndim = dim_0 self.dtype = type_0 except SetInputOrTargetException as e: e.index = index raise e - - def append(self, val:Any): + + def append(self, val: Any): """ :param val: 把该val append到fieldarray。 :return: """ - if (self._is_target or self._is_input) and self._ignore_type is False: + if (self._is_target or self._is_input) and self._ignore_type is False and not self._use_1st_ins_infer_dim_type: type_, dim_ = _get_ele_type_and_dim(val) - if self.dtype!=type_: + if self.dtype != type_: raise AppendToTargetOrInputException(f"Value(type:{type_}) are of different types with " f"previous values(type:{self.dtype}).") - if self._cell_ndim!=dim_: + if self._cell_ndim != dim_: raise AppendToTargetOrInputException(f"Value(dim:{dim_}) are of different dimensions with " f"previous values(dim:{self._cell_ndim}).") self.content.append(val) else: self.content.append(val) - + + def pop(self, index): + """ + 删除该field中index处的元素 + :param int index: 从0开始的数据下标。 + :return: + """ + self.content.pop(index) + def __getitem__(self, indices): return self.get(indices, pad=False) - + def __setitem__(self, idx, val): assert isinstance(idx, int) if (self._is_target or self._is_input) and self.ignore_type is False: # 需要检测类型 type_, dim_ = _get_ele_type_and_dim(val) - if self.dtype!=type_: + if self.dtype != type_: raise RuntimeError(f"Value(type:{type_}) are of different types with " - f"other values(type:{self.dtype}).") - if self._cell_ndim!=dim_: + f"other values(type:{self.dtype}).") + if self._cell_ndim != dim_: raise RuntimeError(f"Value(dim:{dim_}) are of different dimensions with " - f"previous values(dim:{self._cell_ndim}).") + f"previous values(dim:{self._cell_ndim}).") self.content[idx] = val - + def get(self, indices, pad=True): """ 根据给定的indices返回内容 @@ -171,16 +201,16 @@ class FieldArray: return self.content[indices] if self.is_input is False and self.is_target is False: raise RuntimeError("Please specify either is_input or is_target to True for {}".format(self.name)) - + contents = [self.content[i] for i in indices] if self.padder is None or pad is False: return np.array(contents) else: return self.pad(contents) - + def pad(self, contents): return self.padder(contents, field_name=self.name, field_ele_dtype=self.dtype, dim=self._cell_ndim) - + def set_padder(self, padder): """ 设置padder，在这个field进行pad的时候用这个padder进行pad，如果为None则不进行pad。 @@ -192,7 +222,7 @@ class FieldArray: self.padder = deepcopy(padder) else: self.padder = None - + def set_pad_val(self, pad_val): """ 修改padder的pad_val. @@ -202,7 +232,7 @@ class FieldArray: if self.padder is not None: self.padder.set_pad_val(pad_val) return self - + def __len__(self): """ Returns the size of FieldArray. @@ -210,7 +240,7 @@ class FieldArray: :return int length: """ return len(self.content) - + def to(self, other): """ 将other的属性复制给本FieldArray(other必须为FieldArray类型). @@ -220,15 +250,15 @@ class FieldArray: :return: :class:`~fastNLP.FieldArray` """ assert isinstance(other, FieldArray), "Only supports fastNLP.FieldArray type, not {}.".format(type(other)) - + self.ignore_type = other.ignore_type self.is_input = other.is_input self.is_target = other.is_target self.padder = other.padder - + return self - - def split(self, sep:str=None, inplace:bool=True): + + def split(self, sep: str = None, inplace: bool = True): """ 依次对自身的元素使用.split()方法，应该只有当本field的元素为str时，该方法才有用。将返回值 @@ -241,11 +271,11 @@ class FieldArray: try: new_contents.append(cell.split(sep)) except Exception as e: - print(f"Exception happens when process value in index {index}.") + logger.error(f"Exception happens when process value in index {index}.") raise e return self._after_process(new_contents, inplace=inplace) - - def int(self, inplace:bool=True): + + def int(self, inplace: bool = True): """ 将本field中的值调用int(cell). 支持field中内容为以下两种情况(1)['1', '2', ...](即field中每个值为str的)， (2) [['1', '2', ..], ['3', ..], ...](即field中每个值为一个list，list中的值会被依次转换。) @@ -261,10 +291,10 @@ class FieldArray: else: new_contents.append(int(cell)) except Exception as e: - print(f"Exception happens when process value in index {index}.") - print(e) + logger.error(f"Exception happens when process value in index {index}.") + raise e return self._after_process(new_contents, inplace=inplace) - + def float(self, inplace=True): """ 将本field中的值调用float(cell). 支持field中内容为以下两种情况(1)['1', '2', ...](即field中每个值为str的)， @@ -281,10 +311,10 @@ class FieldArray: else: new_contents.append(float(cell)) except Exception as e: - print(f"Exception happens when process value in index {index}.") + logger.error(f"Exception happens when process value in index {index}.") raise e return self._after_process(new_contents, inplace=inplace) - + def bool(self, inplace=True): """ 将本field中的值调用bool(cell). 支持field中内容为以下两种情况(1)['1', '2', ...](即field中每个值为str的)， @@ -301,11 +331,11 @@ class FieldArray: else: new_contents.append(bool(cell)) except Exception as e: - print(f"Exception happens when process value in index {index}.") + logger.error(f"Exception happens when process value in index {index}.") raise e - + return self._after_process(new_contents, inplace=inplace) - + def lower(self, inplace=True): """ 将本field中的值调用cell.lower(). 支持field中内容为以下两种情况(1)['1', '2', ...](即field中每个值为str的)， @@ -322,10 +352,10 @@ class FieldArray: else: new_contents.append(cell.lower()) except Exception as e: - print(f"Exception happens when process value in index {index}.") + logger.error(f"Exception happens when process value in index {index}.") raise e return self._after_process(new_contents, inplace=inplace) - + def upper(self, inplace=True): """ 将本field中的值调用cell.lower(). 支持field中内容为以下两种情况(1)['1', '2', ...](即field中每个值为str的)， @@ -342,10 +372,10 @@ class FieldArray: else: new_contents.append(cell.upper()) except Exception as e: - print(f"Exception happens when process value in index {index}.") + logger.error(f"Exception happens when process value in index {index}.") raise e return self._after_process(new_contents, inplace=inplace) - + def value_count(self): """ 返回该field下不同value的数量。多用于统计label数量 @@ -353,17 +383,18 @@ class FieldArray: :return: Counter, key是label，value是出现次数 """ count = Counter() - + def cum(cell): if _is_iterable(cell) and not isinstance(cell, str): for cell_ in cell: cum(cell_) else: count[cell] += 1 + for cell in self.content: cum(cell) return count - + def _after_process(self, new_contents, inplace): """ 当调用处理函数之后，决定是否要替换field。 @@ -378,14 +409,14 @@ class FieldArray: self.is_input = self.is_input self.is_target = self.is_input except SetInputOrTargetException as e: - print("The newly generated field cannot be set as input or target.") + logger.error("The newly generated field cannot be set as input or target.") raise e return self else: return new_contents -def _get_ele_type_and_dim(cell:Any, dim=0): +def _get_ele_type_and_dim(cell: Any, dim=0): """ 识别cell的类别与dimension的数量 @@ -401,13 +432,13 @@ def _get_ele_type_and_dim(cell:Any, dim=0): elif isinstance(cell, list): dim += 1 res = [_get_ele_type_and_dim(cell_i, dim) for cell_i in cell] - types = set([i for i,j in res]) - dims = set([j for i,j in res]) - if len(types)>1: + types = set([i for i, j in res]) + dims = set([j for i, j in res]) + if len(types) > 1: raise SetInputOrTargetException("Mixed types detected: {}.".format(list(types))) - elif len(types)==0: + elif len(types) == 0: raise SetInputOrTargetException("Empty value encountered.") - if len(dims)>1: + if len(dims) > 1: raise SetInputOrTargetException("Mixed dimension detected: {}.".format(list(dims))) return types.pop(), dims.pop() elif isinstance(cell, torch.Tensor): @@ -418,55 +449,47 @@ def _get_ele_type_and_dim(cell:Any, dim=0): # 否则需要继续往下iterate dim += 1 res = [_get_ele_type_and_dim(cell_i, dim) for cell_i in cell] - types = set([i for i,j in res]) - dims = set([j for i,j in res]) - if len(types)>1: + types = set([i for i, j in res]) + dims = set([j for i, j in res]) + if len(types) > 1: raise SetInputOrTargetException("Mixed types detected: {}.".format(list(types))) - elif len(types)==0: + elif len(types) == 0: raise SetInputOrTargetException("Empty value encountered.") - if len(dims)>1: + if len(dims) > 1: raise SetInputOrTargetException("Mixed dimension detected: {}.".format(list(dims))) return types.pop(), dims.pop() - else: # 包含tuple, set, dict以及其它的类型 + else: # 包含tuple, set, dict以及其它的类型 raise SetInputOrTargetException(f"Cannot process type:{type(cell)}.") -def _is_iterable(value): - # 检查是否是iterable的, duck typing - try: - iter(value) - return True - except BaseException as e: - return False - - class Padder: """ - 别名：:class:`fastNLP.Padder` :class:`fastNLP.core.field.Padder` - 所有padder都需要继承这个类，并覆盖__call__方法。 用于对batch进行padding操作。传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前deepcopy一份。 .. py:function:: __call__(self, contents, field_name, field_ele_dtype): + + """ + + def __init__(self, pad_val=0, **kwargs): + """ - 传入的是List内容。假设有以下的DataSet。 - :param List[Any] contents: 传入的element是inplace的，即直接修改element可能导致数据变化，建议inplace修改之前 deepcopy一份。 :param str, field_name: field的名称。 :param np.int64,np.float64,np.str,None, field_ele_dtype: 该field的内层元素的类型。如果该field的ignore_type为True，该这个值为None。 :return: np.array([padded_element]) - - """ - - def __init__(self, pad_val=0, **kwargs): + """ self.pad_val = pad_val - + def set_pad_val(self, pad_val): self.pad_val = pad_val + def get_pad_val(self): + return self.pad_val + @abstractmethod - def __call__(self, contents, field_name, field_ele_dtype, dim:int): + def __call__(self, contents, field_name, field_ele_dtype, dim: int): """ 传入的是List内容。假设有以下的DataSet。 @@ -512,8 +535,6 @@ class Padder: class AutoPadder(Padder): """ - 别名：:class:`fastNLP.AutoPadder` :class:`fastNLP.core.field.AutoPadder` - 根据contents的数据自动判定是否需要做padding。 1 如果元素类型(元素类型是指field中最里层元素的数据类型, 可以通过FieldArray.dtype查看，比如['This', 'is', ...]的元素类 @@ -533,23 +554,24 @@ class AutoPadder(Padder): 3 其它情况不进行处理，返回一个np.array类型。 """ + def __init__(self, pad_val=0): super().__init__(pad_val=pad_val) - + def __call__(self, contents, field_name, field_ele_dtype, dim): if field_ele_dtype: - if dim>3: + if dim > 3: return np.array(contents) if isinstance(field_ele_dtype, type) and \ (issubclass(field_ele_dtype, np.number) or issubclass(field_ele_dtype, Number)): - if dim==0: + if dim == 0: array = np.array(contents, dtype=field_ele_dtype) - elif dim==1: + elif dim == 1: max_len = max(map(len, contents)) array = np.full((len(contents), max_len), self.pad_val, dtype=field_ele_dtype) for i, content_i in enumerate(contents): array[i, :len(content_i)] = content_i - elif dim==2: + elif dim == 2: max_len = max(map(len, contents)) max_word_len = max([max([len(content_ii) for content_ii in content_i]) for content_i in contents]) @@ -559,20 +581,21 @@ class AutoPadder(Padder): array[i, j, :len(content_ii)] = content_ii else: shape = np.shape(contents) - if len(shape)==4: # 说明各dimension是相同的大小 + if len(shape) == 4: # 说明各dimension是相同的大小 array = np.array(contents, dtype=field_ele_dtype) else: - raise RuntimeError(f"Field:{field_name} has 3 dimensions, every sample should have the same shape.") + raise RuntimeError( + f"Field:{field_name} has 3 dimensions, every sample should have the same shape.") return array elif str(field_ele_dtype).startswith('torch'): - if dim==0: + if dim == 0: tensor = torch.tensor(contents).to(field_ele_dtype) - elif dim==1: + elif dim == 1: max_len = max(map(len, contents)) tensor = torch.full((len(contents), max_len), fill_value=self.pad_val, dtype=field_ele_dtype) for i, content_i in enumerate(contents): - tensor[i, :len(content_i)] = torch.tensor(content_i) - elif dim==2: + tensor[i, :len(content_i)] = content_i.clone().detach() + elif dim == 2: max_len = max(map(len, contents)) max_word_len = max([max([len(content_ii) for content_ii in content_i]) for content_i in contents]) @@ -580,18 +603,21 @@ class AutoPadder(Padder): dtype=field_ele_dtype) for i, content_i in enumerate(contents): for j, content_ii in enumerate(content_i): - tensor[i, j, :len(content_ii)] = torch.tensor(content_ii) + tensor[i, j, :len(content_ii)] = content_ii.clone().detach() else: shapes = set([np.shape(content_i) for content_i in contents]) - if len(shapes)>1: - raise RuntimeError(f"Field:{field_name} has 3 dimensions, every sample should have the same shape.") + if len(shapes) > 1: + raise RuntimeError( + f"Field:{field_name} has 3 dimensions, every sample should have the same shape.") shape = shapes.pop() - if len(shape)==3: - tensor = torch.full([len(contents)]+list(shape), fill_value=self.pad_val, dtype=field_ele_dtype) + if len(shape) == 3: + tensor = torch.full([len(contents)] + list(shape), fill_value=self.pad_val, + dtype=field_ele_dtype) for i, content_i in enumerate(contents): - tensor[i] = torch.tensor(content_i, dtype=field_ele_dtype) + tensor[i] = content_i.clone().detach().to(field_ele_dtype) else: - raise RuntimeError(f"Field:{field_name} has 3 dimensions, every sample should have the same shape.") + raise RuntimeError( + f"Field:{field_name} has 3 dimensions, every sample should have the same shape.") return tensor else: return np.array(contents) # 不进行任何操作 @@ -601,8 +627,6 @@ class AutoPadder(Padder): class EngChar2DPadder(Padder): """ - 别名：:class:`fastNLP.EngChar2DPadder` :class:`fastNLP.core.field.EngChar2DPadder` - 用于为英语执行character级别的2D padding操作。对应的field内容应该类似[['T', 'h', 'i', 's'], ['a'], ['d', 'e', 'm', 'o']]， 但这个Padder只能处理index为int的情况。 @@ -622,7 +646,7 @@ class EngChar2DPadder(Padder): dataset.set_padder('chars', padder) # chars这个field的设置为了EnChar2DPadder """ - + def __init__(self, pad_val=0, pad_length=0): """ :param pad_val: int, pad的位置使用该index @@ -630,9 +654,9 @@ class EngChar2DPadder(Padder): 都pad或截取到该长度. """ super().__init__(pad_val=pad_val) - + self.pad_length = pad_length - + def __call__(self, contents, field_name, field_ele_dtype, dim): """ 期望输入类似于 @@ -651,7 +675,7 @@ class EngChar2DPadder(Padder): raise TypeError('dtype of Field:{} should be np.int64 or np.float64 to do 2D padding, get {}.'.format( field_name, field_ele_dtype )) - assert dim==2, f"Field:{field_name} has {dim}, EngChar2DPadder only supports input with 2 dimensions." + assert dim == 2, f"Field:{field_name} has {dim}, EngChar2DPadder only supports input with 2 dimensions." if self.pad_length < 1: max_char_length = max([max(len(char_lst) for char_lst in word_lst) for word_lst in contents]) else: @@ -659,12 +683,12 @@ class EngChar2DPadder(Padder): max_sent_length = max(len(word_lst) for word_lst in contents) batch_size = len(contents) dtype = type(contents[0][0][0]) - + padded_array = np.full((batch_size, max_sent_length, max_char_length), fill_value=self.pad_val, dtype=dtype) for b_idx, word_lst in enumerate(contents): for c_idx, char_lst in enumerate(word_lst): chars = char_lst[:max_char_length] padded_array[b_idx, c_idx, :len(chars)] = chars - + return padded_array diff --git a/fastNLP/core/instance.py b/fastNLP/core/instance.py index 5408522e..3cf7ab45 100644 --- a/fastNLP/core/instance.py +++ b/fastNLP/core/instance.py @@ -3,15 +3,16 @@ instance 模块实现了Instance 类在fastNLP中对应sample。一个sample可 便于理解的例子可以参考文档 :doc:`fastNLP.core.dataset` 中的表格 """ + __all__ = [ "Instance" ] +from .utils import pretty_table_printer + class Instance(object): """ - 别名：:class:`fastNLP.Instance` :class:`fastNLP.core.instance.Instance` - Instance是fastNLP中对应一个sample的类。每个sample在fastNLP中是一个Instance对象。 Instance一般与 :class:`~fastNLP.DataSet` 一起使用, Instance的初始化如下面的Example所示:: @@ -22,11 +23,11 @@ class Instance(object): >>>ins.add_field("field_3", [3, 3, 3]) >>>ins = Instance(**{'x1': 1, 'x2':np.zeros((3, 4))}) """ - + def __init__(self, **fields): - + self.fields = fields - + def add_field(self, field_name, field): """ 向Instance中增加一个field @@ -35,18 +36,23 @@ class Instance(object): :param Any field: 新增field的内容 """ self.fields[field_name] = field - + + def items(self): + """ + 返回一个迭代器，迭代器返回两个内容，第一个内容是field_name, 第二个内容是field_value + + :return: 一个迭代器 + """ + return self.fields.items() + def __getitem__(self, name): if name in self.fields: return self.fields[name] else: raise KeyError("{} not found".format(name)) - + def __setitem__(self, name, field): return self.add_field(name, field) - + def __repr__(self): - s = '\'' - return "{" + ",\n".join( - "\'" + field_name + "\': " + str(self.fields[field_name]) + \ - f" type={(str(type(self.fields[field_name]))).split(s)[1]}" for field_name in self.fields) + "}" + return str(pretty_table_printer(self)) diff --git a/fastNLP/core/losses.py b/fastNLP/core/losses.py index 1f8923eb..9b32babb 100644 --- a/fastNLP/core/losses.py +++ b/fastNLP/core/losses.py @@ -20,7 +20,6 @@ from collections import defaultdict import torch import torch.nn.functional as F -from ..core.const import Const from .utils import _CheckError from .utils import _CheckRes from .utils import _build_args @@ -28,6 +27,7 @@ from .utils import _check_arg_dict_list from .utils import _check_function_or_method from .utils import _get_func_signature from .utils import seq_len_to_mask +from ..core.const import Const class LossBase(object): @@ -166,8 +166,6 @@ class LossBase(object): class LossFunc(LossBase): """ - 别名：:class:`fastNLP.LossFunc` :class:`fastNLP.core.losses.LossFunc` - 提供给用户使用自定义损失函数的类 :param func: 用户自行定义的损失函数，应当为一个函数或者callable(func)为True的ojbect @@ -199,13 +197,15 @@ class LossFunc(LossBase): class CrossEntropyLoss(LossBase): """ - 别名：:class:`fastNLP.CrossEntropyLoss` :class:`fastNLP.core.losses.CrossEntropyLoss` - 交叉熵损失函数 :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` - :param seq_len: 句子的长度, 长度之外的token不会计算loss。。 + :param seq_len: 句子的长度, 长度之外的token不会计算loss。 + :param int class_in_dim: 在序列标注的场景中，pred可能的shape为(batch_size, max_len, num_classes) + 或(batch_size, num_classes, max_len)， CrossEntropyLoss需要知道哪一维是class的维度以计算loss。如果为-1，就根据pred的第 + 二维是否等于target的第二维来判断是否需要交换pred的第二维和第三维，因为target的第二维是length的维度，如果这一维度上和pred相等， + 那么pred可能第二维也是长度维(存在误判的可能，如果有误判的情况，请显示设置该值)。其它大于0的值则认为该维度是class的维度。 :param padding_idx: padding的index，在计算loss时将忽略target中标号为padding_idx的内容, 可以通过该值代替 传入seq_len. :param str reduction: 支持 `mean` ，`sum` 和 `none` . @@ -216,21 +216,25 @@ class CrossEntropyLoss(LossBase): """ - def __init__(self, pred=None, target=None, seq_len=None, padding_idx=-100, reduction='mean'): + def __init__(self, pred=None, target=None, seq_len=None, class_in_dim=-1, padding_idx=-100, reduction='mean'): super(CrossEntropyLoss, self).__init__() self._init_param_map(pred=pred, target=target, seq_len=seq_len) self.padding_idx = padding_idx assert reduction in ('mean', 'sum', 'none') self.reduction = reduction + self.class_in_dim = class_in_dim def get_loss(self, pred, target, seq_len=None): if pred.dim() > 2: - if pred.size(1) != target.size(1): - pred = pred.transpose(1, 2) + if self.class_in_dim == -1: + if pred.size(1) != target.size(1): # 有可能顺序替换了 + pred = pred.transpose(1, 2) + else: + pred = pred.tranpose(-1, pred) pred = pred.reshape(-1, pred.size(-1)) target = target.reshape(-1) - if seq_len is not None: - mask = seq_len_to_mask(seq_len).reshape(-1).eq(0) + if seq_len is not None and target.dim()>1: + mask = seq_len_to_mask(seq_len, max_len=target.size(1)).reshape(-1).eq(0) target = target.masked_fill(mask, self.padding_idx) return F.cross_entropy(input=pred, target=target, @@ -239,8 +243,6 @@ class CrossEntropyLoss(LossBase): class L1Loss(LossBase): """ - 别名：:class:`fastNLP.L1Loss` :class:`fastNLP.core.losses.L1Loss` - L1损失函数 :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` @@ -261,8 +263,6 @@ class L1Loss(LossBase): class BCELoss(LossBase): """ - 别名：:class:`fastNLP.BCELoss` :class:`fastNLP.core.losses.BCELoss` - 二分类交叉熵损失函数 :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` @@ -282,18 +282,18 @@ class BCELoss(LossBase): class NLLLoss(LossBase): """ - 别名：:class:`fastNLP.NLLLoss` :class:`fastNLP.core.losses.NLLLoss` - 负对数似然损失函数 - - :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` - :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` - :param ignore_idx: ignore的index，在计算loss时将忽略target中标号为ignore_idx的内容, 可以通过该值代替 - 传入seq_len. - :param str reduction: 支持 `mean` ，`sum` 和 `none` . """ def __init__(self, pred=None, target=None, ignore_idx=-100, reduction='mean'): + """ + + :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` + :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` + :param ignore_idx: ignore的index，在计算loss时将忽略target中标号为ignore_idx的内容, 可以通过该值代替 + 传入seq_len. + :param str reduction: 支持 `mean` ，`sum` 和 `none` . + """ super(NLLLoss, self).__init__() self._init_param_map(pred=pred, target=target) assert reduction in ('mean', 'sum', 'none') @@ -306,14 +306,14 @@ class NLLLoss(LossBase): class LossInForward(LossBase): """ - 别名：:class:`fastNLP.LossInForward` :class:`fastNLP.core.losses.LossInForward` - 从forward()函数返回结果中获取loss - - :param str loss_key: 在forward函数中loss的键名，默认为loss """ def __init__(self, loss_key=Const.LOSS): + """ + + :param str loss_key: 在forward函数中loss的键名，默认为loss + """ super().__init__() if not isinstance(loss_key, str): raise TypeError(f"Only str allowed for loss_key, got {type(loss_key)}.") diff --git a/fastNLP/core/metrics.py b/fastNLP/core/metrics.py index f23eab91..72380fd6 100644 --- a/fastNLP/core/metrics.py +++ b/fastNLP/core/metrics.py @@ -10,7 +10,10 @@ __all__ = [ ] import inspect +import warnings +from abc import abstractmethod from collections import defaultdict +from typing import Union import numpy as np import torch @@ -22,7 +25,6 @@ from .utils import _check_arg_dict_list from .utils import _get_func_signature from .utils import seq_len_to_mask from .vocabulary import Vocabulary -from abc import abstractmethod class MetricBase(object): @@ -118,6 +120,7 @@ class MetricBase(object): def __init__(self): self._param_map = {} # key is param in function, value is input param. self._checked = False + self._metric_name = self.__class__.__name__ @property def param_map(self): @@ -135,6 +138,24 @@ class MetricBase(object): @abstractmethod def get_metric(self, reset=True): raise NotImplemented + + def set_metric_name(self, name:str): + """ + 设置metric的名称，默认是Metric的class name. + + :param str name: + :return: self + """ + self._metric_name = name + return self + + def get_metric_name(self): + """ + 返回metric的名称 + + :return: + """ + return self._metric_name def _init_param_map(self, key_map=None, **kwargs): """检查key_map和其他参数map，并将这些映射关系添加到self._param_map @@ -275,17 +296,16 @@ class MetricBase(object): class AccuracyMetric(MetricBase): """ - - 别名：:class:`fastNLP.AccuracyMetric` :class:`fastNLP.core.metrics.AccuracyMetric` - 准确率Metric（其它的Metric参见 :doc:`fastNLP.core.metrics` ） - - :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` - :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` - :param seq_len: 参数映射表中 `seq_len` 的映射关系，None表示映射关系为 `seq_len` -> `seq_len` """ def __init__(self, pred=None, target=None, seq_len=None): + """ + + :param pred: 参数映射表中 `pred` 的映射关系，None表示映射关系为 `pred` -> `pred` + :param target: 参数映射表中 `target` 的映射关系，None表示映射关系为 `target` -> `target` + :param seq_len: 参数映射表中 `seq_len` 的映射关系，None表示映射关系为 `seq_len` -> `seq_len` + """ super().__init__() @@ -318,15 +338,18 @@ class AccuracyMetric(MetricBase): raise TypeError(f"`seq_lens` in {_get_func_signature(self.evaluate)} must be torch.Tensor," f"got {type(seq_len)}.") - if seq_len is not None: - masks = seq_len_to_mask(seq_len=seq_len) + if seq_len is not None and target.dim()>1: + max_len = target.size(1) + masks = seq_len_to_mask(seq_len=seq_len, max_len=max_len) else: masks = None - if pred.size() == target.size(): + if pred.dim() == target.dim(): pass - elif len(pred.size()) == len(target.size()) + 1: + elif pred.dim() == target.dim() + 1: pred = pred.argmax(dim=-1) + if seq_len is None and target.dim()>1: + warnings.warn("You are not passing `seq_len` to exclude pad when calculate accuracy.") else: raise RuntimeError(f"In {_get_func_signature(self.evaluate)}, when pred have " f"size:{pred.size()}, target should have size: {pred.size()} or " @@ -358,6 +381,7 @@ def _bmes_tag_to_spans(tags, ignore_labels=None): """ 给定一个tags的lis，比如['S-song', 'B-singer', 'M-singer', 'E-singer', 'S-moive', 'S-actor']。 返回[('song', (0, 1)), ('singer', (1, 4)), ('moive', (4, 5)), ('actor', (5, 6))] (左闭右开区间) + 也可以是单纯的['S', 'B', 'M', 'E', 'B', 'M', 'M',...]序列 :param tags: List[str], :param ignore_labels: List[str], 在该list中的label将被忽略 @@ -473,10 +497,75 @@ def _bio_tag_to_spans(tags, ignore_labels=None): return [(span[0], (span[1][0], span[1][1] + 1)) for span in spans if span[0] not in ignore_labels] +def _get_encoding_type_from_tag_vocab(tag_vocab:Union[Vocabulary, dict])->str: + """ + 给定Vocabulary自动判断是哪种类型的encoding, 支持判断bmes, bioes, bmeso, bio + + :param tag_vocab: 支持传入tag Vocabulary; 或者传入形如{0:"O", 1:"B-tag1"}，即index在前，tag在后的dict。 + :return: + """ + tag_set = set() + unk_token = '' + pad_token = '' + if isinstance(tag_vocab, Vocabulary): + unk_token = tag_vocab.unknown + pad_token = tag_vocab.padding + tag_vocab = tag_vocab.idx2word + for idx, tag in tag_vocab.items(): + if tag in (unk_token, pad_token): + continue + tag = tag[:1].lower() + tag_set.add(tag) + + bmes_tag_set = set('bmes') + if tag_set == bmes_tag_set: + return 'bmes' + bio_tag_set = set('bio') + if tag_set == bio_tag_set: + return 'bio' + bmeso_tag_set = set('bmeso') + if tag_set == bmeso_tag_set: + return 'bmeso' + bioes_tag_set = set('bioes') + if tag_set == bioes_tag_set: + return 'bioes' + raise RuntimeError("encoding_type cannot be inferred automatically. Only support " + "'bio', 'bmes', 'bmeso', 'bioes' type.") + + +def _check_tag_vocab_and_encoding_type(tag_vocab:Union[Vocabulary, dict], encoding_type:str): + """ + 检查vocab中的tag是否与encoding_type是匹配的 + + :param tag_vocab: 支持传入tag Vocabulary; 或者传入形如{0:"O", 1:"B-tag1"}，即index在前，tag在后的dict。 + :param encoding_type: bio, bmes, bioes, bmeso + :return: + """ + tag_set = set() + unk_token = '' + pad_token = '' + if isinstance(tag_vocab, Vocabulary): + unk_token = tag_vocab.unknown + pad_token = tag_vocab.padding + tag_vocab = tag_vocab.idx2word + for idx, tag in tag_vocab.items(): + if tag in (unk_token, pad_token): + continue + tag = tag[:1].lower() + tag_set.add(tag) + + tags = encoding_type + for tag in tag_set: + assert tag in tags, f"{tag} is not a valid tag in encoding type:{encoding_type}. Please check your " \ + f"encoding_type." + tags = tags.replace(tag, '') # 删除该值 + if tags: # 如果不为空，说明出现了未使用的tag + warnings.warn(f"Tag:{tags} in encoding type:{encoding_type} is not presented in your Vocabulary. Check your " + "encoding_type.") + + class SpanFPreRecMetric(MetricBase): r""" - 别名：:class:`fastNLP.SpanFPreRecMetric` :class:`fastNLP.core.metrics.SpanFPreRecMetric` - 在序列标注问题中，以span的方式计算F, pre, rec. 比如中文Part of speech中，会以character的方式进行标注，句子 `中国在亚洲` 对应的POS可能为(以BMES为例) ['B-NN', 'E-NN', 'S-DET', 'B-NN', 'E-NN']。该metric就是为类似情况下的F1计算。 @@ -499,34 +588,36 @@ class SpanFPreRecMetric(MetricBase): 'rec-label':xxx, ... } - - :param tag_vocab: 标签的 :class:`~fastNLP.Vocabulary` 。支持的标签为"B"(没有label)；或"B-xxx"(xxx为某种label，比如POS中的NN)， - 在解码时，会将相同xxx的认为是同一个label，比如['B-NN', 'E-NN']会被合并为一个'NN'. - :param str pred: 用该key在evaluate()时从传入dict中取出prediction数据。 为None，则使用 `pred` 取数据 - :param str target: 用该key在evaluate()时从传入dict中取出target数据。 为None，则使用 `target` 取数据 - :param str seq_len: 用该key在evaluate()时从传入dict中取出sequence length数据。为None，则使用 `seq_len` 取数据。 - :param str encoding_type: 目前支持bio, bmes, bmeso, bioes - :param list ignore_labels: str 组成的list. 这个list中的class不会被用于计算。例如在POS tagging时传入['NN']，则不会计算'NN'这 - 个label - :param bool only_gross: 是否只计算总的f1, precision, recall的值；如果为False，不仅返回总的f1, pre, rec, 还会返回每个 - label的f1, pre, rec - :param str f_type: `micro` 或 `macro` . `micro` :通过先计算总体的TP，FN和FP的数量，再计算f, precision, recall; `macro` : - 分布计算每个类别的f, precision, recall，然后做平均（各类别f的权重相同） - :param float beta: f_beta分数， :math:`f_{beta} = \frac{(1 + {beta}^{2})*(pre*rec)}{({beta}^{2}*pre + rec)}` . - 常用为beta=0.5, 1, 2. 若为0.5则精确率的权重高于召回率；若为1，则两者平等；若为2，则召回率权重高于精确率。 """ - def __init__(self, tag_vocab, pred=None, target=None, seq_len=None, encoding_type='bio', ignore_labels=None, + def __init__(self, tag_vocab, pred=None, target=None, seq_len=None, encoding_type=None, ignore_labels=None, only_gross=True, f_type='micro', beta=1): - - encoding_type = encoding_type.lower() - + r""" + + :param tag_vocab: 标签的 :class:`~fastNLP.Vocabulary` 。支持的标签为"B"(没有label)；或"B-xxx"(xxx为某种label，比如POS中的NN)， + 在解码时，会将相同xxx的认为是同一个label，比如['B-NN', 'E-NN']会被合并为一个'NN'. + :param str pred: 用该key在evaluate()时从传入dict中取出prediction数据。 为None，则使用 `pred` 取数据 + :param str target: 用该key在evaluate()时从传入dict中取出target数据。 为None，则使用 `target` 取数据 + :param str seq_len: 用该key在evaluate()时从传入dict中取出sequence length数据。为None，则使用 `seq_len` 取数据。 + :param str encoding_type: 目前支持bio, bmes, bmeso, bioes。默认为None，通过tag_vocab自动判断. + :param list ignore_labels: str 组成的list. 这个list中的class不会被用于计算。例如在POS tagging时传入['NN']，则不会计算'NN'个label + :param bool only_gross: 是否只计算总的f1, precision, recall的值；如果为False，不仅返回总的f1, pre, rec, 还会返回每个label的f1, pre, rec + :param str f_type: `micro` 或 `macro` . `micro` :通过先计算总体的TP，FN和FP的数量，再计算f, precision, recall; `macro` : 分布计算每个类别的f, precision, recall，然后做平均（各类别f的权重相同） + :param float beta: f_beta分数， :math:`f_{beta} = \frac{(1 + {beta}^{2})*(pre*rec)}{({beta}^{2}*pre + rec)}` . 常用为 `beta=0.5, 1, 2` 若为0.5则精确率的权重高于召回率；若为1，则两者平等；若为2，则召回率权重高于精确率。 + """ + if not isinstance(tag_vocab, Vocabulary): raise TypeError("tag_vocab can only be fastNLP.Vocabulary, not {}.".format(type(tag_vocab))) if f_type not in ('micro', 'macro'): raise ValueError("f_type only supports `micro` or `macro`', got {}.".format(f_type)) - - self.encoding_type = encoding_type + + if encoding_type: + encoding_type = encoding_type.lower() + _check_tag_vocab_and_encoding_type(tag_vocab, encoding_type) + self.encoding_type = encoding_type + else: + self.encoding_type = _get_encoding_type_from_tag_vocab(tag_vocab) + if self.encoding_type == 'bmes': self.tag_to_span_func = _bmes_tag_to_spans elif self.encoding_type == 'bio': @@ -536,7 +627,7 @@ class SpanFPreRecMetric(MetricBase): elif self.encoding_type == 'bioes': self.tag_to_span_func = _bioes_tag_to_spans else: - raise ValueError("Only support 'bio', 'bmes', 'bmeso' type.") + raise ValueError("Only support 'bio', 'bmes', 'bmeso', 'bioes' type.") self.ignore_labels = ignore_labels self.f_type = f_type @@ -624,7 +715,7 @@ class SpanFPreRecMetric(MetricBase): f, pre, rec = self._compute_f_pre_rec(tp, fn, fp) f_sum += f pre_sum += pre - rec_sum + rec + rec_sum += rec if not self.only_gross and tag != '': # tag!=''防止无tag的情况 f_key = 'f-{}'.format(tag) pre_key = 'pre-{}'.format(tag) @@ -738,24 +829,23 @@ def _pred_topk(y_prob, k=1): class ExtractiveQAMetric(MetricBase): r""" - 别名：:class:`fastNLP.ExtractiveQAMetric` :class:`fastNLP.core.metrics.ExtractiveQAMetric` - 抽取式QA（如SQuAD）的metric. - :param pred1: 参数映射表中 `pred1` 的映射关系，None表示映射关系为 `pred1` -> `pred1` - :param pred2: 参数映射表中 `pred2` 的映射关系，None表示映射关系为 `pred2` -> `pred2` - :param target1: 参数映射表中 `target1` 的映射关系，None表示映射关系为 `target1` -> `target1` - :param target2: 参数映射表中 `target2` 的映射关系，None表示映射关系为 `target2` -> `target2` - :param float beta: f_beta分数， :math:`f_{beta} = \frac{(1 + {beta}^{2})*(pre*rec)}{({beta}^{2}*pre + rec)}` . - 常用为beta=0.5, 1, 2. 若为0.5则精确率的权重高于召回率；若为1，则两者平等；若为2，则召回率权重高于精确率。 - :param bool right_open: right_open为true表示start跟end指针指向一个左闭右开区间，为false表示指向一个左闭右闭区间。 - :param bool print_predict_stat: True则输出预测答案是否为空与正确答案是否为空的统计信息, False则不输出 - """ def __init__(self, pred1=None, pred2=None, target1=None, target2=None, beta=1, right_open=True, print_predict_stat=False): - + r""" + + :param pred1: 参数映射表中 `pred1` 的映射关系，None表示映射关系为 `pred1` -> `pred1` + :param pred2: 参数映射表中 `pred2` 的映射关系，None表示映射关系为 `pred2` -> `pred2` + :param target1: 参数映射表中 `target1` 的映射关系，None表示映射关系为 `target1` -> `target1` + :param target2: 参数映射表中 `target2` 的映射关系，None表示映射关系为 `target2` -> `target2` + :param float beta: f_beta分数， :math:`f_{beta} = \frac{(1 + {beta}^{2})*(pre*rec)}{({beta}^{2}*pre + rec)}` . + 常用为beta=0.5, 1, 2. 若为0.5则精确率的权重高于召回率；若为1，则两者平等；若为2，则召回率权重高于精确率。 + :param bool right_open: right_open为true表示start跟end指针指向一个左闭右开区间，为false表示指向一个左闭右闭区间。 + :param bool print_predict_stat: True则输出预测答案是否为空与正确答案是否为空的统计信息, False则不输出 + """ super(ExtractiveQAMetric, self).__init__() self._init_param_map(pred1=pred1, pred2=pred2, target1=target1, target2=target2) @@ -814,8 +904,8 @@ class ExtractiveQAMetric(MetricBase): if not self.right_open: e += 1 te += 1 - if ts == 0 and te == int(not self.right_open): - if s == 0 and e == int(not self.right_open): + if ts == 0 and te == 1: + if s == 0 and e == 1: self.no_ans_correct += 1 self.no2no += 1 else: diff --git a/fastNLP/core/optimizer.py b/fastNLP/core/optimizer.py index 3036257c..b782cfa6 100644 --- a/fastNLP/core/optimizer.py +++ b/fastNLP/core/optimizer.py @@ -9,21 +9,23 @@ __all__ = [ "AdamW" ] -import torch import math + import torch from torch.optim.optimizer import Optimizer as TorchOptimizer class Optimizer(object): """ - 别名：:class:`fastNLP.Optimizer` :class:`fastNLP.core.optimizer.Optimizer` - - :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models. - :param kwargs: additional parameters. + Optimizer """ def __init__(self, model_params, **kwargs): + """ + + :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models. + :param kwargs: additional parameters. + """ if model_params is not None and not hasattr(model_params, "__next__"): raise RuntimeError("model parameters should be a generator, rather than {}.".format(type(model_params))) self.model_params = model_params @@ -49,7 +51,7 @@ class NullOptimizer(Optimizer): super().__init__(None) def construct_from_pytorch(self, model_params): - pass + return self def __getattr__(self, item): def pass_func(*args, **kwargs): @@ -60,14 +62,15 @@ class NullOptimizer(Optimizer): class SGD(Optimizer): """ - 别名：:class:`fastNLP.SGD` :class:`fastNLP.core.optimizer.SGD` - - :param float lr: learning rate. Default: 0.01 - :param float momentum: momentum. Default: 0 - :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models. + SGD """ def __init__(self, lr=0.001, momentum=0, model_params=None): + """ + :param float lr: learning rate. Default: 0.01 + :param float momentum: momentum. Default: 0 + :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models. + """ if not isinstance(lr, float): raise TypeError("learning rate has to be float.") super(SGD, self).__init__(model_params, lr=lr, momentum=momentum) @@ -82,14 +85,18 @@ class SGD(Optimizer): class Adam(Optimizer): """ - 别名：:class:`fastNLP.Adam` :class:`fastNLP.core.optimizer.Adam` - :param float lr: learning rate - :param float weight_decay: - :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models. """ def __init__(self, lr=0.001, weight_decay=0, betas=(0.9, 0.999), eps=1e-8, amsgrad=False, model_params=None): + """ + + :param float lr: learning rate + :param float weight_decay: + :param eps: + :param amsgrad: + :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models. + """ if not isinstance(lr, float): raise TypeError("learning rate has to be float.") super(Adam, self).__init__(model_params, lr=lr, betas=betas, eps=eps, amsgrad=amsgrad, @@ -105,8 +112,6 @@ class Adam(Optimizer): class AdamW(TorchOptimizer): r""" - 别名：:class:`fastNLP.AdamW` :class:`fastNLP.core.optimizer.AdamW` - 对AdamW的实现，该实现应该会在pytorch更高版本中出现，https://github.com/pytorch/pytorch/pull/21250。这里提前加入 .. todo:: @@ -115,27 +120,28 @@ class AdamW(TorchOptimizer): The original Adam algorithm was proposed in `Adam: A Method for Stochastic Optimization`_. The AdamW variant was proposed in `Decoupled Weight Decay Regularization`_. - :param params (iterable): iterable of parameters to optimize or dicts defining - parameter groups - :param lr (float, optional): learning rate (default: 1e-3) - :param betas (Tuple[float, float], optional): coefficients used for computing - running averages of gradient and its square (default: (0.9, 0.99)) - :param eps (float, optional): term added to the denominator to improve - numerical stability (default: 1e-8) - :param weight_decay (float, optional): weight decay coefficient (default: 1e-2) - algorithm from the paper `On the Convergence of Adam and Beyond`_ - (default: False) - - .. _Adam\: A Method for Stochastic Optimization: - https://arxiv.org/abs/1412.6980 - .. _Decoupled Weight Decay Regularization: - https://arxiv.org/abs/1711.05101 - .. _On the Convergence of Adam and Beyond: - https://openreview.net/forum?id=ryQu7f-RZ + .. _Adam\: A Method for Stochastic Optimization: https://arxiv.org/abs/1412.6980 + + .. _Decoupled Weight Decay Regularization: https://arxiv.org/abs/1711.05101 + + .. _On the Convergence of Adam and Beyond: https://openreview.net/forum?id=ryQu7f-RZ """ def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=1e-2, amsgrad=False): + """ + + :param params (iterable): iterable of parameters to optimize or dicts defining + parameter groups + :param lr (float, optional): learning rate (default: 1e-3) + :param betas (Tuple[float, float], optional): coefficients used for computing + running averages of gradient and its square (default: (0.9, 0.99)) + :param eps (float, optional): term added to the denominator to improve + numerical stability (default: 1e-8) + :param weight_decay (float, optional): weight decay coefficient (default: 1e-2) + algorithm from the paper `On the Convergence of Adam and Beyond`_ + (default: False) + """ if not 0.0 <= lr: raise ValueError("Invalid learning rate: {}".format(lr)) if not 0.0 <= eps: diff --git a/fastNLP/core/predictor.py b/fastNLP/core/predictor.py index 2d6a7380..e4112d5f 100644 --- a/fastNLP/core/predictor.py +++ b/fastNLP/core/predictor.py @@ -1,13 +1,15 @@ -""" - ..todo:: - 检查这个类是否需要 -""" +"""undocumented""" + +__all__ = [ + "Predictor" +] + from collections import defaultdict import torch -from . import DataSetIter from . import DataSet +from . import DataSetIter from . import SequentialSampler from .utils import _build_args, _move_dict_value_to_device, _get_model_device @@ -18,18 +20,20 @@ class Predictor(object): 与测试器（Tester）不同的是，predictor不关心模型性能的评价指标，只做inference。 这是一个fastNLP调用的高级模型包装器。它与Trainer、Tester不共享任何操作。 - - :param torch.nn.Module network: 用来完成预测任务的模型 """ - + def __init__(self, network): + """ + + :param torch.nn.Module network: 用来完成预测任务的模型 + """ if not isinstance(network, torch.nn.Module): raise ValueError( "Only fastNLP.models.BaseModel or torch.nn,Module is allowed, not {}".format(type(network))) self.network = network self.batch_size = 1 self.batch_output = [] - + def predict(self, data: DataSet, seq_len_field_name=None): """用已经训练好的模型进行inference. @@ -41,27 +45,27 @@ class Predictor(object): raise ValueError("Only Dataset class is allowed, not {}.".format(type(data))) if seq_len_field_name is not None and seq_len_field_name not in data.field_arrays: raise ValueError("Field name {} not found in DataSet {}.".format(seq_len_field_name, data)) - + prev_training = self.network.training self.network.eval() network_device = _get_model_device(self.network) batch_output = defaultdict(list) data_iterator = DataSetIter(data, batch_size=self.batch_size, sampler=SequentialSampler(), as_numpy=False) - + if hasattr(self.network, "predict"): predict_func = self.network.predict else: predict_func = self.network.forward - + with torch.no_grad(): for batch_x, _ in data_iterator: _move_dict_value_to_device(batch_x, _, device=network_device) refined_batch_x = _build_args(predict_func, **batch_x) prediction = predict_func(**refined_batch_x) - + if seq_len_field_name is not None: seq_lens = batch_x[seq_len_field_name].tolist() - + for key, value in prediction.items(): value = value.cpu().numpy() if len(value.shape) == 1 or (len(value.shape) == 2 and value.shape[1] == 1): @@ -74,6 +78,6 @@ class Predictor(object): batch_output[key].extend(tmp_batch) else: batch_output[key].append(value) - + self.network.train(prev_training) return batch_output diff --git a/fastNLP/core/sampler.py b/fastNLP/core/sampler.py index d8ba1ad1..6e025688 100644 --- a/fastNLP/core/sampler.py +++ b/fastNLP/core/sampler.py @@ -15,9 +15,6 @@ import numpy as np class Sampler(object): """ - 别名：:class:`fastNLP.Sampler` :class:`fastNLP.core.sampler.Sampler` - - `Sampler` 类的基类. 规定以何种顺序取出data中的元素 子类必须实现 ``__call__`` 方法. 输入 `DataSet` 对象, 返回其中元素的下标序列 @@ -25,16 +22,14 @@ class Sampler(object): def __call__(self, data_set): """ - :param DataSet data_set: `DataSet` 对象, 需要Sample的数据 - :return result: list(int) 其中元素的下标序列, ``data_set`` 中元素会按 ``result`` 中顺序取出 - """ + :param DataSet data_set: `DataSet` 对象, 需要Sample的数据 + :return result: list(int) 其中元素的下标序列, ``data_set`` 中元素会按 ``result`` 中顺序取出 + """ raise NotImplementedError class SequentialSampler(Sampler): """ - 别名：:class:`fastNLP.SequentialSampler` :class:`fastNLP.core.sampler.SequentialSampler` - 顺序取出元素的 `Sampler` """ @@ -45,8 +40,6 @@ class SequentialSampler(Sampler): class RandomSampler(Sampler): """ - 别名：:class:`fastNLP.RandomSampler` :class:`fastNLP.core.sampler.RandomSampler` - 随机化取元素的 `Sampler` """ @@ -57,17 +50,17 @@ class RandomSampler(Sampler): class BucketSampler(Sampler): """ - 别名：:class:`fastNLP.BucketSampler` :class:`fastNLP.core.sampler.BucketSampler` - 带Bucket的 `Random Sampler`. 可以随机地取出长度相似的元素 - - :param int num_buckets: bucket的数量 - :param int batch_size: batch的大小. 默认为None，Trainer在调用BucketSampler时，会将该值正确设置，如果是非Trainer场景使用，需 - 要显示传递该值 - :param str seq_len_field_name: 对应序列长度的 `field` 的名字 """ def __init__(self, num_buckets=10, batch_size=None, seq_len_field_name='seq_len'): + """ + + :param int num_buckets: bucket的数量 + :param int batch_size: batch的大小. 默认为None，Trainer在调用BucketSampler时，会将该值正确设置，如果是非Trainer场景使用，需 + 要显示传递该值 + :param str seq_len_field_name: 对应序列长度的 `field` 的名字 + """ self.num_buckets = num_buckets self.batch_size = batch_size self.seq_len_field_name = seq_len_field_name diff --git a/fastNLP/core/tester.py b/fastNLP/core/tester.py index c1d270d1..d1d5d41e 100644 --- a/fastNLP/core/tester.py +++ b/fastNLP/core/tester.py @@ -32,9 +32,16 @@ Tester在验证进行之前会调用model.eval()提示当前进入了evaluation """ +import time + import torch import torch.nn as nn +try: + from tqdm.auto import tqdm +except: + from .utils import _pseudo_tqdm as tqdm + from .batch import BatchIter, DataSetIter from .dataset import DataSet from .metrics import _prepare_metrics @@ -47,7 +54,9 @@ from .utils import _get_func_signature from .utils import _get_model_device from .utils import _move_model_to_device from ._parallel_utils import _data_parallel_wrapper +from ._parallel_utils import _model_contains_inner_module from functools import partial +from ._logger import logger __all__ = [ "Tester" @@ -56,36 +65,35 @@ __all__ = [ class Tester(object): """ - 别名：:class:`fastNLP.Tester` :class:`fastNLP.core.tester.Tester` - Tester是在提供数据，模型以及metric的情况下进行性能测试的类。需要传入模型，数据以及metric进行验证。 - - :param ~fastNLP.DataSet data: 需要测试的数据集 - :param torch.nn.module model: 使用的模型 - :param ~fastNLP.core.metrics.MetricBase,List[~fastNLP.core.metrics.MetricBase] metrics: 测试时使用的metrics - :param int batch_size: evaluation时使用的batch_size有多大。 - :param str,int,torch.device,list(int) device: 将模型load到哪个设备。默认为None，即Trainer不对模型 - 的计算位置进行管理。支持以下的输入: - - 1. str: ['cpu', 'cuda', 'cuda:0', 'cuda:1', ...] 依次为'cpu'中, 可见的第一个GPU中,可见的第一个GPU中,可见的第二个GPU中; - - 2. torch.device：将模型装载到torch.device上。 - - 3. int: 将使用device_id为该值的gpu进行训练 - - 4. list(int)：如果多于1个device，将使用torch.nn.DataParallel包裹model, 并使用传入的device。 - - 5. None. 为None则不对模型进行任何处理，如果传入的model为torch.nn.DataParallel该值必须为None。 - - 如果模型是通过predict()进行预测的话，那么将不能使用多卡(DataParallel)进行验证，只会使用第一张卡上的模型。 - :param int verbose: 如果为0不输出任何信息; 如果为1，打印出验证结果。 """ - def __init__(self, data, model, metrics, batch_size=16, num_workers=0, device=None, verbose=1): - super(Tester, self).__init__() + def __init__(self, data, model, metrics, batch_size=16, num_workers=0, device=None, verbose=1, use_tqdm=True): + """ - if not isinstance(data, DataSet): - raise TypeError(f"The type of data must be `fastNLP.DataSet`, got `{type(data)}`.") + :param ~fastNLP.DataSet data: 需要测试的数据集 + :param torch.nn.module model: 使用的模型 + :param ~fastNLP.core.metrics.MetricBase,List[~fastNLP.core.metrics.MetricBase] metrics: 测试时使用的metrics + :param int batch_size: evaluation时使用的batch_size有多大。 + :param str,int,torch.device,list(int) device: 将模型load到哪个设备。默认为None，即Trainer不对模型 + 的计算位置进行管理。支持以下的输入: + + 1. str: ['cpu', 'cuda', 'cuda:0', 'cuda:1', ...] 依次为'cpu'中, 可见的第一个GPU中,可见的第一个GPU中,可见的第二个GPU中; + + 2. torch.device：将模型装载到torch.device上。 + + 3. int: 将使用device_id为该值的gpu进行训练 + + 4. list(int)：如果多于1个device，将使用torch.nn.DataParallel包裹model, 并使用传入的device。 + + 5. None. 为None则不对模型进行任何处理，如果传入的model为torch.nn.DataParallel该值必须为None。 + + 如果模型是通过predict()进行预测的话，那么将不能使用多卡(DataParallel)进行验证，只会使用第一张卡上的模型。 + :param int verbose: 如果为0不输出任何信息; 如果为1，打印出验证结果。 + :param bool use_tqdm: 是否使用tqdm来显示测试进度; 如果为False，则不会显示任何内容。 + """ + super(Tester, self).__init__() + if not isinstance(model, nn.Module): raise TypeError(f"The type of model must be `torch.nn.Module`, got `{type(model)}`.") @@ -95,6 +103,8 @@ class Tester(object): self._model = _move_model_to_device(model, device=device) self.batch_size = batch_size self.verbose = verbose + self.use_tqdm = use_tqdm + self.logger = logger if isinstance(data, DataSet): self.data_iterator = DataSetIter( @@ -106,19 +116,22 @@ class Tester(object): # check predict if (hasattr(self._model, 'predict') and callable(self._model.predict)) or \ - (isinstance(self._model, nn.DataParallel) and hasattr(self._model.module, 'predict') and - callable(self._model.module.predict)): + (_model_contains_inner_module(self._model) and hasattr(self._model.module, 'predict') and + callable(self._model.module.predict)): if isinstance(self._model, nn.DataParallel): self._predict_func_wrapper = partial(_data_parallel_wrapper('predict', self._model.device_ids, self._model.output_device), network=self._model.module) + self._predict_func = self._model.module.predict # 用于匹配参数 + elif isinstance(self._model, nn.parallel.DistributedDataParallel): self._predict_func = self._model.module.predict + self._predict_func_wrapper = self._model.module.predict # 用于调用 else: self._predict_func = self._model.predict self._predict_func_wrapper = self._model.predict else: - if isinstance(self._model, nn.DataParallel): + if _model_contains_inner_module(model): self._predict_func_wrapper = self._model.forward self._predict_func = self._model.module.forward else: @@ -126,10 +139,9 @@ class Tester(object): self._predict_func_wrapper = self._model.forward def test(self): - """开始进行验证，并返回验证结果。 + r"""开始进行验证，并返回验证结果。 - :return Dict[Dict] : dict的二层嵌套结构，dict的第一层是metric的名称; 第二层是这个metric的指标。 - 一个AccuracyMetric的例子为{'AccuracyMetric': {'acc': 1.0}}。 + :return Dict[Dict]: dict的二层嵌套结构，dict的第一层是metric的名称; 第二层是这个metric的指标。一个AccuracyMetric的例子为{'AccuracyMetric': {'acc': 1.0}}。 """ # turn on the testing mode; clean up the history self._model_device = _get_model_device(self._model) @@ -139,21 +151,39 @@ class Tester(object): eval_results = {} try: with torch.no_grad(): - for batch_x, batch_y in data_iterator: - _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) - pred_dict = self._data_forward(self._predict_func, batch_x) - if not isinstance(pred_dict, dict): - raise TypeError(f"The return value of {_get_func_signature(self._predict_func)} " - f"must be `dict`, got {type(pred_dict)}.") + if not self.use_tqdm: + from .utils import _pseudo_tqdm as inner_tqdm + else: + inner_tqdm = tqdm + with inner_tqdm(total=len(data_iterator), leave=False, dynamic_ncols=True) as pbar: + pbar.set_description_str(desc="Test") + + start_time = time.time() + + for batch_x, batch_y in data_iterator: + _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) + pred_dict = self._data_forward(self._predict_func, batch_x) + if not isinstance(pred_dict, dict): + raise TypeError(f"The return value of {_get_func_signature(self._predict_func)} " + f"must be `dict`, got {type(pred_dict)}.") + for metric in self.metrics: + metric(pred_dict, batch_y) + + if self.use_tqdm: + pbar.update() + for metric in self.metrics: - metric(pred_dict, batch_y) - for metric in self.metrics: - eval_result = metric.get_metric() - if not isinstance(eval_result, dict): - raise TypeError(f"The return value of {_get_func_signature(metric.get_metric)} must be " - f"`dict`, got {type(eval_result)}") - metric_name = metric.__class__.__name__ - eval_results[metric_name] = eval_result + eval_result = metric.get_metric() + if not isinstance(eval_result, dict): + raise TypeError(f"The return value of {_get_func_signature(metric.get_metric)} must be " + f"`dict`, got {type(eval_result)}") + metric_name = metric.get_metric_name() + eval_results[metric_name] = eval_result + pbar.close() + end_time = time.time() + test_str = f'Evaluate data in {round(end_time - start_time, 2)} seconds!' + # pbar.write(test_str) + self.logger.info(test_str) except _CheckError as e: prev_func_signature = _get_func_signature(self._predict_func) _check_loss_evaluate(prev_func_signature=prev_func_signature, func_signature=e.func_signature, @@ -161,7 +191,7 @@ class Tester(object): dataset=self.data, check_level=0) if self.verbose >= 1: - print("[tester] \n{}".format(self._format_eval_results(eval_results))) + logger.info("[tester] \n{}".format(self._format_eval_results(eval_results))) self._mode(network, is_test=False) return eval_results diff --git a/fastNLP/core/trainer.py b/fastNLP/core/trainer.py index 671e2736..a2c3b1f7 100644 --- a/fastNLP/core/trainer.py +++ b/fastNLP/core/trainer.py @@ -336,7 +336,7 @@ except: import warnings from .batch import DataSetIter, BatchIter -from .callback import CallbackManager, CallbackException +from .callback import CallbackManager, CallbackException, Callback from .dataset import DataSet from .losses import _prepare_losser from .metrics import _prepare_metrics @@ -352,12 +352,11 @@ from .utils import _move_dict_value_to_device from .utils import _get_func_signature from .utils import _get_model_device from .utils import _move_model_to_device - +from ._parallel_utils import _model_contains_inner_module +from ._logger import logger class Trainer(object): """ - 别名：:class:`fastNLP.Trainer` :class:`fastNLP.core.trainer.Trainer` - Trainer在fastNLP中用于组织单任务的训练过程，可以避免用户在不同训练任务中重复撰写 (1) epoch循环; (2) 将数据分成不同的Batch; @@ -366,87 +365,84 @@ class Trainer(object): (5) 保存获得更好验证性能的模型等。 详细的介绍参见 :doc:`fastNLP.core.trainer` - - :param train_data: 训练集， :class:`~fastNLP.DataSet` 类型。 - :param nn.modules model: 待训练的模型 - :param optimizer: `torch.optim.Optimizer` 优化器。如果为None，则Trainer使用默认的Adam(model.parameters(), lr=4e-3)这个优化器 - :param int batch_size: 训练和验证的时候的batch大小。 - :param loss: 使用的 :class:`~fastNLP.core.losses.LossBase` 对象。当为None时，默认使用 :class:`~fastNLP.LossInForward` - :param sampler: Batch数据生成的顺序， :class:`~fastNLP.Sampler` 类型。如果为None，默认使用 :class:`~fastNLP.RandomSampler` - :param drop_last: 如果最后一个batch没有正好为batch_size这么多数据，就扔掉最后一个batch - :param num_workers: int, 有多少个线程来进行数据pad处理。 - :param update_every: int, 多少步更新一次梯度。用于希望累计梯度的场景，比如需要128的batch_size, 但是直接设为128 - 会导致内存不足，通过设置batch_size=32, update_every=4达到目的。当optimizer为None时，该参数无效。 - :param int n_epochs: 需要优化迭代多少次。 - :param int print_every: 多少次反向传播更新tqdm显示的loss; 如果use_tqdm=False, 则多少次反向传播打印loss。 - :param dev_data: 用于做验证的DataSet， :class:`~fastNLP.DataSet` 类型。 - :param metrics: 验证的评估函数。可以只使用一个 :class:`Metric` ， - 也可以使用多个 :class:`Metric` ，通过列表传入。 - 如验证时取得了更好的验证结果(如果有多个Metric，以列表中第一个Metric为准)，且save_path不为None， - 则保存当前模型。Metric种类详见 :doc:`metrics模块 ` 。仅在传入dev_data时有效。 - :param str,None metric_key: :class:`Metric` 有时会有多个指标， - 比如 :class:`~fastNLP.core.metrics.SpanFPreRecMetric` 中包含了'f', 'pre', 'rec'。此时需 - 要指定以哪个指标为准。另外有些指标是越小效果越好，比如语言模型的困惑度，这种情况下，在key前面增加一个'-'来表 - 明验证时，值越小越好(比如: "-ppl")。仅在传入dev_data时有效。 - :param int validate_every: 多少个step在验证集上验证一次; 如果为-1，则每个epoch结束验证一次。仅在传入dev_data时有效。 - :param str,None save_path: 将模型保存路径。如果为None，则不保存模型。如果dev_data为None，则保存最后一次迭代的模型。 - 保存的时候不仅保存了参数，还保存了模型结构。即便使用DataParallel，这里也只保存模型。 - :param bool use_tqdm: 是否使用tqdm来显示训练进度; 如果为False，则将loss打印在终端中。 - :param str,int,torch.device,list(int) device: 将模型load到哪个设备。默认为None，即Trainer不对模型 - 的计算位置进行管理。支持以下的输入: - - 1. str: ['cpu', 'cuda', 'cuda:0', 'cuda:1', ...] 依次为'cpu'中, 可见的第一个GPU中, 可见的第一个GPU中, - 可见的第二个GPU中; - - 2. torch.device：将模型装载到torch.device上。 - - 3. int: 将使用device_id为该值的gpu进行训练 - - 4. list(int)：如果多于1个device，将使用torch.nn.DataParallel包裹model, 并使用传入的device。 - - 5. None. 为None则不对模型进行任何处理，如果传入的model为torch.nn.DataParallel该值必须为None。 - - 已知可能会出现的问题：Adagrad优化器可能无法正常使用这个参数，请手动管理模型位置。 - - :param list(callbacks) callbacks: 用于在train过程中起调节作用的回调函数。比如early stop，negative sampling等可以 - 通过callback机制实现。 可使用的callback参见 :doc:`callback模块 ` - :param int check_code_level: 模型检查等级. -1: 不进行检查; 0: 仅出现错误时停止; 1: 如果有field没有被使用， - 报告警告信息; 2: 有任何field没有被使用都报错. 检查的原理是通过使用很小的batch(默认2个sample)来运行代码，但是 - 这个过程理论上不会修改任何参数，只是会检查能否运行。但如果(1)模型中存在将batch_size写为某个固定值的情况； - (2)模型中存在累加前向计算次数的，可能会多计算1次。以上情况建议将check_code_level设置为-1。 """ def __init__(self, train_data, model, optimizer=None, loss=None, batch_size=32, sampler=None, drop_last=False, update_every=1, num_workers=0, n_epochs=10, print_every=5, dev_data=None, metrics=None, metric_key=None, - validate_every=-1, save_path=None, use_tqdm=True, device=None, prefetch=False, - callbacks=None, check_code_level=0): - if prefetch and num_workers==0: - num_workers = 1 - if prefetch: - warnings.warn("prefetch is deprecated, will be removed in version 0.5.0, please use num_workers instead.") - + validate_every=-1, save_path=None, use_tqdm=True, device=None, + callbacks=None, check_code_level=0, **kwargs): + """ + + :param train_data: 训练集， :class:`~fastNLP.DataSet` 类型。 + :param nn.modules model: 待训练的模型 + :param optimizer: `torch.optim.Optimizer` 优化器。如果为None，则Trainer使用默认的Adam(model.parameters(), lr=4e-3)这个优化器 + :param int batch_size: 训练和验证的时候的batch大小。 + :param loss: 使用的 :class:`~fastNLP.core.losses.LossBase` 对象。当为None时，默认使用 :class:`~fastNLP.LossInForward` + :param sampler: Batch数据生成的顺序， :class:`~fastNLP.Sampler` 类型。如果为None，默认使用 :class:`~fastNLP.RandomSampler` + :param drop_last: 如果最后一个batch没有正好为batch_size这么多数据，就扔掉最后一个batch + :param num_workers: int, 有多少个线程来进行数据pad处理。 + :param update_every: int, 多少步更新一次梯度。用于希望累计梯度的场景，比如需要128的batch_size, 但是直接设为128 + 会导致内存不足，通过设置batch_size=32, update_every=4达到目的。当optimizer为None时，该参数无效。 + :param int n_epochs: 需要优化迭代多少次。 + :param int print_every: 多少次反向传播更新tqdm显示的loss; 如果use_tqdm=False, 则多少次反向传播打印loss。 + :param dev_data: 用于做验证的DataSet， :class:`~fastNLP.DataSet` 类型。 + :param metrics: 验证的评估函数。可以只使用一个 :class:`Metric` ， + 也可以使用多个 :class:`Metric` ，通过列表传入。 + 如验证时取得了更好的验证结果(如果有多个Metric，以列表中第一个Metric为准)，且save_path不为None， + 则保存当前模型。Metric种类详见 :doc:`metrics模块 ` 。仅在传入dev_data时有效。 + :param str,None metric_key: :class:`Metric` 有时会有多个指标， + 比如 :class:`~fastNLP.core.metrics.SpanFPreRecMetric` 中包含了'f', 'pre', 'rec'。此时需 + 要指定以哪个指标为准。另外有些指标是越小效果越好，比如语言模型的困惑度，这种情况下，在key前面增加一个'-'来表 + 明验证时，值越小越好(比如: "-ppl")。仅在传入dev_data时有效。 + :param int validate_every: 多少个step在验证集上验证一次; 如果为-1，则每个epoch结束验证一次。仅在传入dev_data时有效。 + :param str,None save_path: 将模型保存路径，如果路径不存在，将自动创建文件夹。如果为None，则不保存模型。如果dev_data为None，则保存 + 最后一次迭代的模型。保存的时候不仅保存了参数，还保存了模型结构。即便使用DataParallel，这里也只保存模型。 + :param bool use_tqdm: 是否使用tqdm来显示训练进度; 如果为False，则将loss打印在终端中。 + :param str,int,torch.device,list(int) device: 将模型load到哪个设备。默认为None，即Trainer不对模型 + 的计算位置进行管理。支持以下的输入: + + 1. str: ['cpu', 'cuda', 'cuda:0', 'cuda:1', ...] 依次为'cpu'中, 可见的第一个GPU中, 可见的第一个GPU中, + 可见的第二个GPU中; + + 2. torch.device：将模型装载到torch.device上。 + + 3. int: 将使用device_id为该值的gpu进行训练 + + 4. list(int)：如果多于1个device，将使用torch.nn.DataParallel包裹model, 并使用传入的device。 + + 5. None. 为None则不对模型进行任何处理，如果传入的model为torch.nn.DataParallel该值必须为None。 + + 已知可能会出现的问题：Adagrad优化器可能无法正常使用这个参数，请手动管理模型位置。 + + :param list(callbacks) callbacks: 用于在train过程中起调节作用的回调函数。比如early stop，negative sampling等可以 + 通过callback机制实现。 可使用的callback参见 :doc:`callback模块 ` + :param int check_code_level: 模型检查等级. -1: 不进行检查; 0: 仅出现错误时停止; 1: 如果有field没有被使用， + 报告警告信息; 2: 有任何field没有被使用都报错. 检查的原理是通过使用很小的batch(默认2个sample)来运行代码，但是 + 这个过程理论上不会修改任何参数，只是会检查能否运行。但如果(1)模型中存在将batch_size写为某个固定值的情况； + (2)模型中存在累加前向计算次数的，可能会多计算1次。以上情况建议将check_code_level设置为-1。 + """ super(Trainer, self).__init__() if not isinstance(model, nn.Module): raise TypeError(f"The type of model must be torch.nn.Module, got {type(model)}.") - + # check metrics and dev_data if (not metrics) and dev_data is not None: raise ValueError("No metric for dev_data evaluation.") if metrics and (dev_data is None): raise ValueError("No dev_data for evaluations, pass dev_data or set metrics to None. ") - + # check update every assert update_every >= 1, "update_every must be no less than 1." self.update_every = int(update_every) - + # check save_path if not (save_path is None or isinstance(save_path, str)): raise ValueError("save_path can only be None or `str`.") # prepare evaluate metrics = _prepare_metrics(metrics) - + # parse metric_key # increase_better is True. It means the exp result gets better if the indicator increases. # It is true by default. @@ -458,30 +454,69 @@ class Trainer(object): self.metric_key = None # prepare loss losser = _prepare_losser(loss) - - # sampler check - if sampler is not None and not isinstance(sampler, Sampler): - raise ValueError("The type of sampler should be fastNLP.BaseSampler, got {}.".format(type(sampler))) - if sampler is None: - sampler = RandomSampler() - elif hasattr(sampler, 'set_batch_size'): - sampler.set_batch_size(batch_size) + if isinstance(train_data, BatchIter): + if sampler is not None: + warnings.warn("sampler is ignored when train_data is a BatchIter.") + if num_workers>0: + warnings.warn("num_workers is ignored when train_data is BatchIter.") + if drop_last: + warnings.warn("drop_last is ignored when train_data is BatchIter.") + + if isinstance(model, nn.parallel.DistributedDataParallel): # 如果是分布式的 + # device为None + if device is not None: + warnings.warn("device is ignored when model is nn.parallel.DistributedDataParallel.") + device = None + # Sampler要是分布式的 + if sampler is None: + sampler = torch.utils.data.DistributedSampler(train_data) + elif not isinstance(sampler, torch.utils.data.DistributedSampler): + raise TypeError("When using nn.parallel.DistributedDataParallel, " + "sampler must be None or torch.utils.data.DistributedSampler.") + # 不能保存模型 + if save_path: + raise RuntimeError("Saving model in Distributed situation is not allowed right now.") + else: + # sampler check + if sampler is not None and not isinstance(sampler, (Sampler, torch.utils.data.Sampler)): + raise ValueError(f"The type of sampler should be fastNLP.BaseSampler or pytorch's Sampler, got {type(sampler)}") + if sampler is None: + sampler = RandomSampler() + elif hasattr(sampler, 'set_batch_size'): + sampler.set_batch_size(batch_size) if isinstance(train_data, DataSet): self.data_iterator = DataSetIter( dataset=train_data, batch_size=batch_size, num_workers=num_workers, sampler=sampler, drop_last=drop_last) elif isinstance(train_data, BatchIter): self.data_iterator = train_data + train_data = train_data.dataset else: raise TypeError("train_data type {} not support".format(type(train_data))) - if check_code_level > -1 and isinstance(self.data_iterator, DataSetIter): - _check_code(dataset=train_data, model=model, losser=losser, metrics=metrics, dev_data=dev_data, - metric_key=self.metric_key, check_level=check_code_level, - batch_size=min(batch_size, DEFAULT_CHECK_BATCH_SIZE)) - # _check_code 是 fastNLP 帮助你检查代码是否正确的方法 。如果你在错误栈中看到这行注释，请认真检查你的代码 self.model = _move_model_to_device(model, device=device) + if _model_contains_inner_module(self.model): + self._forward_func = self.model.module.forward + else: + self._forward_func = self.model.forward + if check_code_level > -1: + # _check_code 是 fastNLP 帮助你检查代码是否正确的方法 。如果你在错误栈中看到这行注释，请认真检查你的field名与模型的输入 + # 名是否匹配 + dev_dataset = dev_data + if isinstance(dev_data, BatchIter): + dev_dataset = None + warnings.warn("dev_data is of BatchIter type, ignore validation checking.") + check_batch_size = min(batch_size, DEFAULT_CHECK_BATCH_SIZE) + if isinstance(self.model, nn.DataParallel): + _num_devices = len(self.model.device_ids) + if batch_size//_num_devices>1: # 如果多卡是每个卡可以分多个数据的，则用每个卡给两个sample + check_batch_size = max(len(self.model.device_ids)*2, check_batch_size) + else: + check_batch_size = max(len(self.model.device_ids), check_batch_size) + _check_code(dataset=train_data, model=self.model, losser=losser, forward_func=self._forward_func, metrics=metrics, + dev_data=dev_dataset, metric_key=self.metric_key, check_level=check_code_level, + batch_size=check_batch_size) self.train_data = train_data self.dev_data = dev_data # If None, No validation. @@ -496,8 +531,7 @@ class Trainer(object): self.best_dev_epoch = None self.best_dev_step = None self.best_dev_perf = None - self.n_steps = (len(self.train_data) // self.batch_size + int( - len(self.train_data) % self.batch_size != 0)) * int(drop_last==0) * self.n_epochs + self.n_steps = len(self.data_iterator) * self.n_epochs if isinstance(optimizer, torch.optim.Optimizer): self.optimizer = optimizer @@ -507,22 +541,32 @@ class Trainer(object): self.optimizer = torch.optim.Adam(self.model.parameters(), lr=4e-3) else: raise TypeError("optimizer can only be torch.optim.Optimizer type, not {}.".format(type(optimizer))) - + + self.logger = logger + self.use_tqdm = use_tqdm + if 'test_use_tqdm' in kwargs: + self.test_use_tqdm = kwargs.get('test_use_tqdm') + else: + self.test_use_tqdm = self.use_tqdm self.pbar = None self.print_every = abs(self.print_every) - + self.kwargs = kwargs if self.dev_data is not None: self.tester = Tester(model=self.model, data=self.dev_data, metrics=self.metrics, - batch_size=self.batch_size, + batch_size=kwargs.get("dev_batch_size", self.batch_size), device=None, # 由上面的部分处理device - verbose=0) - + verbose=0, + use_tqdm=self.test_use_tqdm) + self.step = 0 self.start_time = None # start timestamp - + + if isinstance(callbacks, Callback): + callbacks = [callbacks] + self.callback_manager = CallbackManager(env={"trainer": self}, callbacks=callbacks) @@ -548,7 +592,7 @@ class Trainer(object): """ results = {} if self.n_epochs <= 0: - print(f"training epoch is {self.n_epochs}, nothing was done.") + self.logger.info(f"training epoch is {self.n_epochs}, nothing was done.") results['seconds'] = 0. return results try: @@ -557,8 +601,8 @@ class Trainer(object): self._load_best_model = load_best_model self.start_time = str(datetime.now().strftime('%Y-%m-%d-%H-%M-%S')) start_time = time.time() - print("training epochs started " + self.start_time, flush=True) - + self.logger.info("training epochs started " + self.start_time) + try: self.callback_manager.on_train_begin() self._train() @@ -571,11 +615,11 @@ class Trainer(object): raise e elif on_exception == 'raise': raise e - + if self.dev_data is not None and self.best_dev_perf is not None: - print( - "\nIn Epoch:{}/Step:{}, got best dev performance:".format(self.best_dev_epoch, self.best_dev_step) + - self.tester._format_eval_results(self.best_dev_perf), ) + self.logger.info( + "\nIn Epoch:{}/Step:{}, got best dev performance:".format(self.best_dev_epoch, self.best_dev_step)) + self.logger.info(self.tester._format_eval_results(self.best_dev_perf)) results['best_eval'] = self.best_dev_perf results['best_epoch'] = self.best_dev_epoch results['best_step'] = self.best_dev_step @@ -583,27 +627,23 @@ class Trainer(object): model_name = "best_" + "_".join([self.model.__class__.__name__, self.metric_key, self.start_time]) load_succeed = self._load_model(self.model, model_name) if load_succeed: - print("Reloaded the best model.") + self.logger.info("Reloaded the best model.") else: - print("Fail to reload best model.") + self.logger.info("Fail to reload best model.") finally: pass results['seconds'] = round(time.time() - start_time, 2) - + return results - + def _train(self): if not self.use_tqdm: - from fastNLP.core.utils import _pseudo_tqdm as inner_tqdm + from .utils import _pseudo_tqdm as inner_tqdm else: inner_tqdm = tqdm self.step = 0 self.epoch = 0 start = time.time() - if isinstance(self.model, nn.DataParallel): - self._forward_func = self.model.module.forward - else: - self._forward_func = self.model.forward with inner_tqdm(total=self.n_steps, postfix='loss:{0:<6.5f}', leave=False, dynamic_ncols=True) as pbar: self.pbar = pbar avg_loss = 0 @@ -621,21 +661,21 @@ class Trainer(object): # negative sampling; replace unknown; re-weight batch_y self.callback_manager.on_batch_begin(batch_x, batch_y, indices) prediction = self._data_forward(self.model, batch_x) - + # edit prediction self.callback_manager.on_loss_begin(batch_y, prediction) loss = self._compute_loss(prediction, batch_y).mean() avg_loss += loss.item() loss = loss / self.update_every - + # Is loss NaN or inf? requires_grad = False self.callback_manager.on_backward_begin(loss) self._grad_backward(loss) self.callback_manager.on_backward_end() - + self._update() self.callback_manager.on_step_end() - + if self.step % self.print_every == 0: avg_loss = float(avg_loss) / self.print_every if self.use_tqdm: @@ -649,36 +689,36 @@ class Trainer(object): pbar.set_postfix_str(print_output) avg_loss = 0 self.callback_manager.on_batch_end() - + if ((self.validate_every > 0 and self.step % self.validate_every == 0) or (self.validate_every < 0 and self.step % len(data_iterator) == 0)) \ and self.dev_data is not None: eval_res = self._do_validation(epoch=epoch, step=self.step) - eval_str = "Evaluation at Epoch {}/{}. Step:{}/{}. ".format(epoch, self.n_epochs, self.step, - self.n_steps) + \ - self.tester._format_eval_results(eval_res) - pbar.write(eval_str + '\n') - + eval_str = "Evaluation on dev at Epoch {}/{}. Step:{}/{}: ".format(epoch, self.n_epochs, self.step, + self.n_steps) + # pbar.write(eval_str + '\n') + self.logger.info(eval_str) + self.logger.info(self.tester._format_eval_results(eval_res)+'\n') # ================= mini-batch end ==================== # - + # lr decay; early stopping self.callback_manager.on_epoch_end() # =============== epochs end =================== # pbar.close() self.pbar = None # ============ tqdm end ============== # - + def _do_validation(self, epoch, step): self.callback_manager.on_valid_begin() res = self.tester.test() - + is_better_eval = False if self._better_eval_result(res): if self.save_path is not None: self._save_model(self.model, "best_" + "_".join([self.model.__class__.__name__, self.metric_key, self.start_time])) elif self._load_best_model: - self._best_model_states = {name: param.cpu().clone() for name, param in self.model.named_parameters()} + self._best_model_states = {name: param.cpu().clone() for name, param in self.model.state_dict().items()} self.best_dev_perf = res self.best_dev_epoch = epoch self.best_dev_step = step @@ -686,7 +726,7 @@ class Trainer(object): # get validation results; adjust optimizer self.callback_manager.on_valid_end(res, self.metric_key, self.optimizer, is_better_eval) return res - + def _mode(self, model, is_test=False): """Train mode or Test mode. This is for PyTorch currently. @@ -698,14 +738,14 @@ class Trainer(object): model.eval() else: model.train() - + def _update(self): """Perform weight update on a model. """ if self.step % self.update_every == 0: self.optimizer.step() - + def _data_forward(self, network, x): x = _build_args(self._forward_func, **x) y = network(**x) @@ -713,7 +753,7 @@ class Trainer(object): raise TypeError( f"The return value of {_get_func_signature(self._forward_func)} should be dict, got {type(y)}.") return y - + def _grad_backward(self, loss): """Compute gradient with link rules. @@ -724,7 +764,7 @@ class Trainer(object): if (self.step-1) % self.update_every == 0: self.model.zero_grad() loss.backward() - + def _compute_loss(self, predict, truth): """Compute loss given prediction and ground truth. @@ -733,7 +773,7 @@ class Trainer(object): :return: a scalar """ return self.losser(predict, truth) - + def _save_model(self, model, model_name, only_param=False): """ 存储不含有显卡信息的state_dict或model :param model: @@ -745,7 +785,7 @@ class Trainer(object): model_path = os.path.join(self.save_path, model_name) if not os.path.exists(self.save_path): os.makedirs(self.save_path, exist_ok=True) - if isinstance(model, nn.DataParallel): + if _model_contains_inner_module(model): model = model.module if only_param: state_dict = model.state_dict() @@ -756,7 +796,7 @@ class Trainer(object): model.cpu() torch.save(model, model_path) model.to(self._model_device) - + def _load_model(self, model, model_name, only_param=False): # 返回bool值指示是否成功reload模型 if self.save_path is not None: @@ -765,7 +805,7 @@ class Trainer(object): states = torch.load(model_path) else: states = torch.load(model_path).state_dict() - if isinstance(model, nn.DataParallel): + if _model_contains_inner_module(model): model.module.load_state_dict(states) else: model.load_state_dict(states) @@ -774,7 +814,7 @@ class Trainer(object): else: return False return True - + def _better_eval_result(self, metrics): """Check if the current epoch yields better validation results. @@ -789,17 +829,20 @@ class Trainer(object): self.best_metric_indicator = indicator_val else: if self.increase_better is True: - if indicator_val > self.best_metric_indicator: + if indicator_val >= self.best_metric_indicator: self.best_metric_indicator = indicator_val else: is_better = False else: - if indicator_val < self.best_metric_indicator: + if indicator_val <= self.best_metric_indicator: self.best_metric_indicator = indicator_val else: is_better = False return is_better + @property + def is_master(self): + return True DEFAULT_CHECK_BATCH_SIZE = 2 DEFAULT_CHECK_NUM_BATCH = 2 @@ -821,14 +864,15 @@ def _get_value_info(_dict): strs.append(_str) return strs + from numbers import Number from .batch import _to_tensor -def _check_code(dataset, model, losser, metrics, batch_size=DEFAULT_CHECK_BATCH_SIZE, - dev_data=None, metric_key=None, - check_level=0): + + +def _check_code(dataset, model, losser, metrics, forward_func, batch_size=DEFAULT_CHECK_BATCH_SIZE, + dev_data=None, metric_key=None, check_level=0): # check get_loss 方法 - model_devcie = _get_model_device(model=model) - + model_device = _get_model_device(model=model) def _iter(): start_idx = 0 while start_idx -1, "device can only be non-negative integer" assert torch.cuda.device_count() > device, "Only has {} gpus, cannot use device {}.".format( @@ -312,7 +269,7 @@ def _get_model_device(model): """ # TODO 这个函数存在一定的风险，因为同一个模型可能存在某些parameter不在显卡中，比如BertEmbedding. 或者跨显卡 assert isinstance(model, nn.Module) - + parameters = list(model.parameters()) if len(parameters) == 0: return None @@ -352,7 +309,6 @@ def _map_args(maps: dict, **kwargs): output.update({name: val}) for keys in maps.keys(): if keys not in output.keys(): - # TODO: add UNUSED warning. pass return output @@ -473,10 +429,10 @@ def _move_dict_value_to_device(*args, device: torch.device, non_blocking=False): """ if not torch.cuda.is_available(): return - + if not isinstance(device, torch.device): raise TypeError(f"device must be `torch.device`, got `{type(device)}`") - + for arg in args: if isinstance(arg, dict): for key, value in arg.items(): @@ -491,10 +447,10 @@ class _CheckError(Exception): _CheckError. Used in losses.LossBase, metrics.MetricBase. """ - + def __init__(self, check_res: _CheckRes, func_signature: str): errs = [f'Problems occurred when calling `{func_signature}`'] - + if check_res.varargs: errs.append(f"\tvarargs: {check_res.varargs}(Does not support pass positional arguments, please delete it)") if check_res.missing: @@ -503,9 +459,9 @@ class _CheckError(Exception): errs.append(f"\tduplicated param: {check_res.duplicated}") if check_res.unused: errs.append(f"\tunused param: {check_res.unused}") - + Exception.__init__(self, '\n'.join(errs)) - + self.check_res = check_res self.func_signature = func_signature @@ -525,7 +481,7 @@ def _check_loss_evaluate(prev_func_signature: str, func_signature: str, check_re # if check_res.varargs: # errs.append(f"\tvarargs: *{check_res.varargs}") # suggestions.append(f"Does not support pass positional arguments, please delete *{check_res.varargs}.") - + if check_res.unused: for _unused in check_res.unused: if _unused in target_dict: @@ -536,7 +492,7 @@ def _check_loss_evaluate(prev_func_signature: str, func_signature: str, check_re unuseds.append(f"\tunused field: {_unused_field}") if _unused_param: unuseds.append(f"\tunused param: {_unused_param}") # output from predict or forward - + module_name = func_signature.split('.')[0] if check_res.missing: errs.append(f"\tmissing param: {check_res.missing}") @@ -557,7 +513,7 @@ def _check_loss_evaluate(prev_func_signature: str, func_signature: str, check_re mapped_missing.append(_miss) else: unmapped_missing.append(_miss) - + for _miss in mapped_missing + unmapped_missing: if _miss in dataset: suggestions.append(f"Set `{_miss}` as target.") @@ -570,29 +526,17 @@ def _check_loss_evaluate(prev_func_signature: str, func_signature: str, check_re else: _tmp = f'Provide `{_miss}` in DataSet or output of {prev_func_signature}.' suggestions.append(_tmp) - # for _miss in unmapped_missing: - # if _miss in dataset: - # suggestions.append(f"Set `{_miss}` as target.") - # else: - # _tmp = '' - # if check_res.unused: - # _tmp = f"Specify your assignment for `{input_func_map.get(_miss, _miss)}` when initialize {module_name}." - # if _tmp: - # _tmp += f' Or provide `{_miss}` in DataSet or output of {prev_func_signature}.' - # else: - # _tmp = f'Provide `{_miss}` in output of {prev_func_signature} or DataSet.' - # suggestions.append(_tmp) - + if check_res.duplicated: errs.append(f"\tduplicated param: {check_res.duplicated}.") suggestions.append(f"Delete {check_res.duplicated} in the output of " f"{prev_func_signature} or do not set {check_res.duplicated} as targets. ") - + if len(errs) > 0: errs.extend(unuseds) elif check_level == STRICT_CHECK_LEVEL: errs.extend(unuseds) - + if len(errs) > 0: errs.insert(0, f'Problems occurred when calling {func_signature}') sugg_str = "" @@ -619,11 +563,11 @@ def _check_loss_evaluate(prev_func_signature: str, func_signature: str, check_re def _check_forward_error(forward_func, batch_x, dataset, check_level): check_res = _check_arg_dict_list(forward_func, batch_x) func_signature = _get_func_signature(forward_func) - + errs = [] suggestions = [] _unused = [] - + # if check_res.varargs: # errs.append(f"\tvarargs: {check_res.varargs}") # suggestions.append(f"Does not support pass positional arguments, please delete *{check_res.varargs}.") @@ -644,14 +588,14 @@ def _check_forward_error(forward_func, batch_x, dataset, check_level): # _tmp += f"Or you might find it in `unused field:`, you can use DataSet.rename_field() to " \ # f"rename the field in `unused field:`." suggestions.append(_tmp) - + if check_res.unused: _unused = [f"\tunused field: {check_res.unused}"] if len(errs) > 0: errs.extend(_unused) elif check_level == STRICT_CHECK_LEVEL: errs.extend(_unused) - + if len(errs) > 0: errs.insert(0, f'Problems occurred when calling {func_signature}') sugg_str = "" @@ -699,7 +643,7 @@ def seq_len_to_mask(seq_len, max_len=None): max_len = int(max_len) if max_len else int(seq_len.max()) broad_cast_seq_len = np.tile(np.arange(max_len), (len(seq_len), 1)) mask = broad_cast_seq_len < seq_len.reshape(-1, 1) - + elif isinstance(seq_len, torch.Tensor): assert seq_len.dim() == 1, f"seq_len can only have one dimension, got {seq_len.dim() == 1}." batch_size = seq_len.size(0) @@ -708,7 +652,7 @@ def seq_len_to_mask(seq_len, max_len=None): mask = broad_cast_seq_len.lt(seq_len.unsqueeze(1)) else: raise TypeError("Only support 1-d numpy.ndarray or 1-d torch.Tensor.") - + return mask @@ -716,25 +660,25 @@ class _pseudo_tqdm: """ 当无法引入tqdm，或者Trainer中设置use_tqdm为false的时候，用该方法打印数据 """ - + def __init__(self, **kwargs): - pass - + self.logger = logger + def write(self, info): - print(info) - + self.logger.info(info) + def set_postfix_str(self, info): - print(info) - + self.logger.info(info) + def __getattr__(self, item): def pass_func(*args, **kwargs): pass - + return pass_func - + def __enter__(self): return self - + def __exit__(self, exc_type, exc_val, exc_tb): del self @@ -788,3 +732,76 @@ def iob2bioes(tags: List[str]) -> List[str]: else: raise TypeError("Invalid IOB format.") return new_tags + + +def _is_iterable(value): + # 检查是否是iterable的, duck typing + try: + iter(value) + return True + except BaseException as e: + return False + + +def get_seq_len(words, pad_value=0): + """ + 给定batch_size x max_len的words矩阵，返回句子长度 + + :param words: batch_size x max_len + :return: (batch_size,) + """ + mask = words.ne(pad_value) + return mask.sum(dim=-1) + + +def pretty_table_printer(dataset_or_ins) -> PrettyTable: + """ + :param dataset_or_ins: 传入一个dataSet或者instance + ins = Instance(field_1=[1, 1, 1], field_2=[2, 2, 2], field_3=["a", "b", "c"]) + +-----------+-----------+-----------------+ + | field_1 | field_2 | field_3 | + +-----------+-----------+-----------------+ + | [1, 1, 1] | [2, 2, 2] | ['a', 'b', 'c'] | + +-----------+-----------+-----------------+ + :return: 以 pretty table的形式返回根据terminal大小进行自动截断 + """ + x = PrettyTable() + try: + sz = os.get_terminal_size() + column = sz.columns + row = sz.lines + except OSError: + column = 144 + row = 11 + if type(dataset_or_ins).__name__ == "DataSet": + x.field_names = list(dataset_or_ins.field_arrays.keys()) + c_size = len(x.field_names) + for ins in dataset_or_ins: + x.add_row([sub_column(ins[k], column, c_size, k) for k in x.field_names]) + row -= 1 + if row < 0: + x.add_row(["..." for _ in range(c_size)]) + break + elif type(dataset_or_ins).__name__ == "Instance": + x.field_names = list(dataset_or_ins.fields.keys()) + c_size = len(x.field_names) + x.add_row([sub_column(dataset_or_ins[k], column, c_size, k) for k in x.field_names]) + + else: + raise Exception("only accept DataSet and Instance") + return x + + +def sub_column(string: str, c: int, c_size: int, title: str) -> str: + """ + :param string: 要被截断的字符串 + :param c: 命令行列数 + :param c_size: instance或dataset field数 + :param title: 列名 + :return: 对一个过长的列进行截断的结果 + """ + avg = max(int(c / c_size), len(title)) + string = str(string) + if len(string) > avg: + string = string[:(avg - 3)] + "..." + return string diff --git a/fastNLP/core/vocabulary.py b/fastNLP/core/vocabulary.py index 9ce59a8c..6d530eb6 100644 --- a/fastNLP/core/vocabulary.py +++ b/fastNLP/core/vocabulary.py @@ -1,14 +1,21 @@ +""" +.. todo:: + doc +""" + __all__ = [ "Vocabulary", "VocabularyOption", ] +from collections import Counter +from functools import partial from functools import wraps -from collections import Counter, defaultdict + +from ._logger import logger from .dataset import DataSet from .utils import Option -from functools import partial -import numpy as np +from .utils import _is_iterable class VocabularyOption(Option): @@ -32,7 +39,7 @@ def _check_build_vocab(func): @wraps(func) # to solve missing docstring def _wrapper(self, *args, **kwargs): - if self.word2idx is None or self.rebuild is True: + if self._word2idx is None or self.rebuild is True: self.build_vocab() return func(self, *args, **kwargs) @@ -49,8 +56,8 @@ def _check_build_status(func): if self.rebuild is False: self.rebuild = True if self.max_size is not None and len(self.word_count) >= self.max_size: - print("[Warning] Vocabulary has reached the max size {} when calling {} method. " - "Adding more words may cause unexpected behaviour of Vocabulary. ".format( + logger.info("[Warning] Vocabulary has reached the max size {} when calling {} method. " + "Adding more words may cause unexpected behaviour of Vocabulary. ".format( self.max_size, func.__name__)) return func(self, *args, **kwargs) @@ -59,8 +66,6 @@ def _check_build_status(func): class Vocabulary(object): """ - 别名：:class:`fastNLP.Vocabulary` :class:`fastNLP.core.vocabulary.Vocabulary` - 用于构建, 存储和使用 `str` 到 `int` 的一一映射:: vocab = Vocabulary() @@ -68,32 +73,52 @@ class Vocabulary(object): vocab.update(word_list) vocab["word"] # str to int vocab.to_word(5) # int to str - - :param int max_size: `Vocabulary` 的最大大小, 即能存储词的最大数量 - 若为 ``None`` , 则不限制大小. Default: ``None`` - :param int min_freq: 能被记录下的词在文本中的最小出现频率, 应大于或等于 1. - 若小于该频率, 词语将被视为 `unknown`. 若为 ``None`` , 所有文本中的词都被记录. Default: ``None`` - :param str optional padding: padding的字符. 如果设置为 ``None`` , - 则vocabulary中不考虑padding, 也不计入词表大小，为 ``None`` 的情况多在为label建立Vocabulary的情况. - Default: '' - :param str optional unknown: unknown的字符，所有未被记录的词在转为 `int` 时将被视为unknown. - 如果设置为 ``None`` ,则vocabulary中不考虑unknow, 也不计入词表大小. - 为 ``None`` 的情况多在为label建立Vocabulary的情况. - Default: '' """ def __init__(self, max_size=None, min_freq=None, padding='', unknown=''): + """ + + :param int max_size: `Vocabulary` 的最大大小, 即能存储词的最大数量 + 若为 ``None`` , 则不限制大小. Default: ``None`` + :param int min_freq: 能被记录下的词在文本中的最小出现频率, 应大于或等于 1. + 若小于该频率, 词语将被视为 `unknown`. 若为 ``None`` , 所有文本中的词都被记录. Default: ``None`` + :param str optional padding: padding的字符. 如果设置为 ``None`` , + 则vocabulary中不考虑padding, 也不计入词表大小，为 ``None`` 的情况多在为label建立Vocabulary的情况. + Default: '' + :param str optional unknown: unknown的字符，所有未被记录的词在转为 `int` 时将被视为unknown. + 如果设置为 ``None`` ,则vocabulary中不考虑unknow, 也不计入词表大小. + 为 ``None`` 的情况多在为label建立Vocabulary的情况. + Default: '' + """ self.max_size = max_size self.min_freq = min_freq self.word_count = Counter() self.unknown = unknown self.padding = padding - self.word2idx = None - self.idx2word = None + self._word2idx = None + self._idx2word = None self.rebuild = True # 用于承载不需要单独创建entry的词语，具体见from_dataset()方法 self._no_create_word = Counter() - + + @property + @_check_build_vocab + def word2idx(self): + return self._word2idx + + @word2idx.setter + def word2idx(self, value): + self._word2idx = value + + @property + @_check_build_vocab + def idx2word(self): + return self._idx2word + + @idx2word.setter + def idx2word(self, value): + self._word2idx = value + @_check_build_status def update(self, word_lst, no_create_entry=False): """依次增加序列中词在词典中的出现频率 @@ -131,11 +156,11 @@ class Vocabulary(object): """ 在新加入word时，检查_no_create_word的设置。 - :param str, List[str] word: + :param str List[str] word: :param bool no_create_entry: :return: """ - if isinstance(word, str): + if isinstance(word, str) or not _is_iterable(word): word = [word] for w in word: if no_create_entry and self.word_count.get(w, 0) == self._no_create_word.get(w, 0): @@ -180,36 +205,36 @@ class Vocabulary(object): 但已经记录在词典中的词, 不会改变对应的 `int` """ - if self.word2idx is None: - self.word2idx = {} + if self._word2idx is None: + self._word2idx = {} if self.padding is not None: - self.word2idx[self.padding] = len(self.word2idx) + self._word2idx[self.padding] = len(self._word2idx) if self.unknown is not None: - self.word2idx[self.unknown] = len(self.word2idx) + self._word2idx[self.unknown] = len(self._word2idx) max_size = min(self.max_size, len(self.word_count)) if self.max_size else None words = self.word_count.most_common(max_size) if self.min_freq is not None: words = filter(lambda kv: kv[1] >= self.min_freq, words) - if self.word2idx is not None: - words = filter(lambda kv: kv[0] not in self.word2idx, words) - start_idx = len(self.word2idx) - self.word2idx.update({w: i + start_idx for i, (w, _) in enumerate(words)}) + if self._word2idx is not None: + words = filter(lambda kv: kv[0] not in self._word2idx, words) + start_idx = len(self._word2idx) + self._word2idx.update({w: i + start_idx for i, (w, _) in enumerate(words)}) self.build_reverse_vocab() self.rebuild = False return self - + def build_reverse_vocab(self): """ 基于 `word to index` dict, 构建 `index to word` dict. """ - self.idx2word = {i: w for w, i in self.word2idx.items()} + self._idx2word = {i: w for w, i in self._word2idx.items()} return self @_check_build_vocab def __len__(self): - return len(self.word2idx) + return len(self._word2idx) @_check_build_vocab def __contains__(self, item): @@ -219,7 +244,7 @@ class Vocabulary(object): :param item: the word :return: True or False """ - return item in self.word2idx + return item in self._word2idx def has_word(self, w): """ @@ -241,12 +266,12 @@ class Vocabulary(object): vocab[w] """ - if w in self.word2idx: - return self.word2idx[w] + if w in self._word2idx: + return self._word2idx[w] if self.unknown is not None: - return self.word2idx[self.unknown] + return self._word2idx[self.unknown] else: - raise ValueError("word {} not in vocabulary".format(w)) + raise ValueError("word `{}` not in vocabulary".format(w)) @_check_build_vocab def index_dataset(self, *datasets, field_name, new_field_name=None): @@ -257,37 +282,47 @@ class Vocabulary(object): vocab.index_dataset(train_data, dev_data, test_data, field_name='words') :param ~fastNLP.DataSet,List[~fastNLP.DataSet] datasets: 需要转index的一个或多个数据集 - :param str field_name: 需要转index的field, 若有多个 DataSet, 每个DataSet都必须有此 field. - 目前仅支持 ``str`` , ``List[str]`` , ``List[List[str]]`` - :param str new_field_name: 保存结果的field_name. 若为 ``None`` , 将覆盖原field. - Default: ``None`` + :param list,str field_name: 需要转index的field, 若有多个 DataSet, 每个DataSet都必须有此 field. + 目前支持 ``str`` , ``List[str]`` + :param list,str new_field_name: 保存结果的field_name. 若为 ``None`` , 将覆盖原field. + Default: ``None``. """ - def index_instance(ins): + def index_instance(field): """ 有几种情况, str, 1d-list, 2d-list :param ins: :return: """ - field = ins[field_name] - if isinstance(field, str): + if isinstance(field, str) or not _is_iterable(field): return self.to_index(field) - elif isinstance(field, list): - if not isinstance(field[0], list): + else: + if isinstance(field[0], str) or not _is_iterable(field[0]): return [self.to_index(w) for w in field] else: - if isinstance(field[0][0], list): + if not isinstance(field[0][0], str) and _is_iterable(field[0][0]): raise RuntimeError("Only support field with 2 dimensions.") return [[self.to_index(c) for c in w] for w in field] - if new_field_name is None: - new_field_name = field_name + new_field_name = new_field_name or field_name + + if type(new_field_name) == type(field_name): + if isinstance(new_field_name, list): + assert len(new_field_name) == len(field_name), "new_field_name should have same number elements with " \ + "field_name." + elif isinstance(new_field_name, str): + field_name = [field_name] + new_field_name = [new_field_name] + else: + raise TypeError("field_name and new_field_name can only be str or List[str].") + for idx, dataset in enumerate(datasets): if isinstance(dataset, DataSet): try: - dataset.apply(index_instance, new_field_name=new_field_name) + for f_n, n_f_n in zip(field_name, new_field_name): + dataset.apply_field(index_instance, field_name=f_n, new_field_name=n_f_n) except Exception as e: - print("When processing the `{}` dataset, the following error occurred.".format(idx)) + logger.info("When processing the `{}` dataset, the following error occurred.".format(idx)) raise e else: raise RuntimeError("Only DataSet type is allowed.") @@ -306,9 +341,8 @@ class Vocabulary(object): :param ~fastNLP.DataSet,List[~fastNLP.DataSet] datasets: 需要转index的一个或多个数据集 :param str,List[str] field_name: 可为 ``str`` 或 ``List[str]`` . - 构建词典所使用的 field(s), 支持一个或多个field - 若有多个 DataSet, 每个DataSet都必须有这些field. - 目前仅支持的field结构: ``str`` , ``List[str]`` , ``list[List[str]]`` + 构建词典所使用的 field(s), 支持一个或多个field，若有多个 DataSet, 每个DataSet都必须有这些field. 目前支持的field结构 + : ``str`` , ``List[str]`` :param no_create_entry_dataset: 可以传入DataSet, List[DataSet]或者None(默认)，该选项用在接下来的模型会使用pretrain 的embedding(包括glove, word2vec, elmo与bert)且会finetune的情况。如果仅使用来自于train的数据建立vocabulary，会导致test与dev 中的数据无法充分利用到来自于预训练embedding的信息，所以在建立词表的时候将test与dev考虑进来会使得最终的结果更好。 @@ -326,14 +360,14 @@ class Vocabulary(object): def construct_vocab(ins, no_create_entry=False): for fn in field_name: field = ins[fn] - if isinstance(field, str): + if isinstance(field, str) or not _is_iterable(field): self.add_word(field, no_create_entry=no_create_entry) - elif isinstance(field, (list, np.ndarray)): - if not isinstance(field[0], (list, np.ndarray)): + else: + if isinstance(field[0], str) or not _is_iterable(field[0]): for word in field: self.add_word(word, no_create_entry=no_create_entry) else: - if isinstance(field[0][0], (list, np.ndarray)): + if not isinstance(field[0][0], str) and _is_iterable(field[0][0]): raise RuntimeError("Only support field with 2 dimensions.") for words in field: for word in words: @@ -343,8 +377,8 @@ class Vocabulary(object): if isinstance(dataset, DataSet): try: dataset.apply(construct_vocab) - except Exception as e: - print("When processing the `{}` dataset, the following error occurred.".format(idx)) + except BaseException as e: + logger.error("When processing the `{}` dataset, the following error occurred:".format(idx)) raise e else: raise TypeError("Only DataSet type is allowed.") @@ -370,7 +404,7 @@ class Vocabulary(object): def to_index(self, w): """ - 将词转为数字. 若词不再词典中被记录, 将视为 unknown, 若 ``unknown=None`` , 将抛出``ValueError``:: + 将词转为数字. 若词不再词典中被记录, 将视为 unknown, 若 ``unknown=None`` , 将抛出 ``ValueError`` :: index = vocab.to_index('abc') # equals to @@ -389,7 +423,7 @@ class Vocabulary(object): """ if self.unknown is None: return None - return self.word2idx[self.unknown] + return self._word2idx[self.unknown] @property @_check_build_vocab @@ -399,7 +433,7 @@ class Vocabulary(object): """ if self.padding is None: return None - return self.word2idx[self.padding] + return self._word2idx[self.padding] @_check_build_vocab def to_word(self, idx): @@ -409,7 +443,7 @@ class Vocabulary(object): :param int idx: the index :return str word: the word """ - return self.idx2word[idx] + return self._idx2word[idx] def clear(self): """ @@ -418,8 +452,8 @@ class Vocabulary(object): :return: """ self.word_count.clear() - self.word2idx = None - self.idx2word = None + self._word2idx = None + self._idx2word = None self.rebuild = True self._no_create_word.clear() return self @@ -430,8 +464,8 @@ class Vocabulary(object): """ len(self) # make sure vocab has been built state = self.__dict__.copy() - # no need to pickle idx2word as it can be constructed from word2idx - del state['idx2word'] + # no need to pickle _idx2word as it can be constructed from _word2idx + del state['_idx2word'] return state def __setstate__(self, state): @@ -446,5 +480,5 @@ class Vocabulary(object): @_check_build_vocab def __iter__(self): - for word, index in self.word2idx.items(): + for word, index in self._word2idx.items(): yield word, index diff --git a/fastNLP/doc_utils.py b/fastNLP/doc_utils.py new file mode 100644 index 00000000..5f293d3f --- /dev/null +++ b/fastNLP/doc_utils.py @@ -0,0 +1,27 @@ +"""undocumented""" + +__all__ = [] + +import inspect +import sys + + +def doc_process(m): + for name, obj in inspect.getmembers(m): + if inspect.isclass(obj) or inspect.isfunction(obj): + if obj.__module__ != m.__name__: + if obj.__doc__ is None: + # print(name, obj.__doc__) + pass + else: + module_name = obj.__module__ + while 1: + defined_m = sys.modules[module_name] + if "undocumented" not in defined_m.__doc__ and name in defined_m.__all__: + obj.__doc__ = r"别名 :class:`" + m.__name__ + "." + name + "`" \ + + " :class:`" + module_name + "." + name + "`\n" + obj.__doc__ + break + module_name = ".".join(module_name.split('.')[:-1]) + if module_name == m.__name__: + # print(name, ": not found defined doc.") + break diff --git a/fastNLP/embeddings/__init__.py b/fastNLP/embeddings/__init__.py index 2bfb2960..ea99154e 100644 --- a/fastNLP/embeddings/__init__.py +++ b/fastNLP/embeddings/__init__.py @@ -7,20 +7,25 @@ torch.FloatTensor。所有的embedding都可以使用 `self.num_embedding` 获 __all__ = [ "Embedding", + "TokenEmbedding", "StaticEmbedding", "ElmoEmbedding", "BertEmbedding", + "BertWordPieceEncoder", "StackEmbedding", "LSTMCharEmbedding", "CNNCharEmbedding", - "get_embeddings" + "get_embeddings", ] - -from .embedding import Embedding +from .embedding import Embedding, TokenEmbedding from .static_embedding import StaticEmbedding from .elmo_embedding import ElmoEmbedding -from .bert_embedding import BertEmbedding +from .bert_embedding import BertEmbedding, BertWordPieceEncoder from .char_embedding import CNNCharEmbedding, LSTMCharEmbedding from .stack_embedding import StackEmbedding -from .utils import get_embeddings \ No newline at end of file +from .utils import get_embeddings + +import sys +from ..doc_utils import doc_process +doc_process(sys.modules[__name__]) \ No newline at end of file diff --git a/fastNLP/embeddings/bert_embedding.py b/fastNLP/embeddings/bert_embedding.py index aa72898a..05351cbd 100644 --- a/fastNLP/embeddings/bert_embedding.py +++ b/fastNLP/embeddings/bert_embedding.py @@ -1,3 +1,12 @@ +""" +.. todo:: + doc +""" + +__all__ = [ + "BertEmbedding", + "BertWordPieceEncoder" +] import os import collections @@ -8,15 +17,15 @@ import numpy as np from itertools import chain from ..core.vocabulary import Vocabulary -from ..io.file_utils import _get_base_url, cached_path, PRETRAINED_BERT_MODEL_DIR +from ..io.file_utils import PRETRAINED_BERT_MODEL_DIR from ..modules.encoder.bert import _WordPieceBertModel, BertModel, BertTokenizer from .contextual_embedding import ContextualEmbedding +import warnings +from ..core import logger class BertEmbedding(ContextualEmbedding): """ - 别名：:class:`fastNLP.embeddings.BertEmbedding` :class:`fastNLP.embeddings.bert_embedding.BertEmbedding` - 使用BERT对words进行编码的Embedding。建议将输入的words长度限制在430以内，而不要使用512(根据预训练模型参数，可能有变化)。这是由于 预训练的bert模型长度限制为512个token，而因为输入的word是未进行word piece分割的(word piece的分割有BertEmbedding在输入word 时切分)，在分割之后长度可能会超过最大长度限制。 @@ -27,6 +36,7 @@ class BertEmbedding(ContextualEmbedding): >>> import torch >>> from fastNLP import Vocabulary + >>> from fastNLP.embeddings import BertEmbedding >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) >>> embed = BertEmbedding(vocab, model_dir_or_name='en-base-uncased', requires_grad=False, layers='4,-2,-1') >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) @@ -37,8 +47,8 @@ class BertEmbedding(ContextualEmbedding): :param ~fastNLP.Vocabulary vocab: 词表 :param str model_dir_or_name: 模型所在目录或者模型的名称。当传入模型所在目录时，目录中应该包含一个词表文件(以.txt作为后缀名), 权重文件(以.bin作为文件后缀名), 配置文件(以.json作为后缀名)。 - :param str layers: 输出embedding表示来自于哪些层，不同层的结果按照layers中的顺序在最后一维concat起来。以','隔开层数，可以以负数 - 去索引倒数几层。 + :param str layers: 输出embedding表示来自于哪些层，不同层的结果按照layers中的顺序在最后一维concat起来。以','隔开层数，层的序号是 + 从0开始，可以以负数去索引倒数几层。 :param str pool_method: 因为在bert中，每个word会被表示为多个word pieces, 当获取一个word的表示的时候，怎样从它的word pieces 中计算得到它对应的表示。支持 ``last`` , ``first`` , ``avg`` , ``max``。 :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 @@ -46,34 +56,40 @@ class BertEmbedding(ContextualEmbedding): :param bool include_cls_sep: bool，在bert计算句子的表示的时候，需要在前面加上[CLS]和[SEP], 是否在结果中保留这两个内容。 这样 会使得word embedding的结果比输入的结果长两个token。如果该值为True，则在使用 :class::StackEmbedding 可能会与其它类型的 embedding长度不匹配。 + :param bool pooled_cls: 返回的[CLS]是否使用预训练中的BertPool映射一下，仅在include_cls_sep时有效。如果下游任务只取[CLS]做预测， + 一般该值为True。 :param bool requires_grad: 是否需要gradient以更新Bert的权重。 + :param bool auto_truncate: 当句子words拆分为word pieces长度超过bert最大允许长度(一般为512), 自动截掉拆分后的超过510个 + word pieces后的内容，并将第512个word piece置为[SEP]。超过长度的部分的encode结果直接全部置零。一般仅有只使用[CLS] + 来进行分类的任务将auto_truncate置为True。 """ - def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en-base-uncased', layers: str='-1', - pool_method: str='first', word_dropout=0, dropout=0, requires_grad: bool=False, - include_cls_sep: bool=False): + + def __init__(self, vocab: Vocabulary, model_dir_or_name: str = 'en-base-uncased', layers: str = '-1', + pool_method: str = 'first', word_dropout=0, dropout=0, include_cls_sep: bool = False, + pooled_cls=True, requires_grad: bool = True, auto_truncate: bool = False): super(BertEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - # 根据model_dir_or_name检查是否存在并下载 if model_dir_or_name.lower() in PRETRAINED_BERT_MODEL_DIR: - PRETRAIN_URL = _get_base_url('bert') - model_name = PRETRAINED_BERT_MODEL_DIR[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): - model_dir = model_dir_or_name - else: - raise ValueError(f"Cannot recognize {model_dir_or_name}.") - - self.model = _WordBertModel(model_dir=model_dir, vocab=vocab, layers=layers, - pool_method=pool_method, include_cls_sep=include_cls_sep) - + if 'cn' in model_dir_or_name.lower() and pool_method not in ('first', 'last'): + logger.warn("For Chinese bert, pooled_method should choose from 'first', 'last' in order to achieve" + " faster speed.") + warnings.warn("For Chinese bert, pooled_method should choose from 'first', 'last' in order to achieve" + " faster speed.") + + self._word_sep_index = None + if '[SEP]' in vocab: + self._word_sep_index = vocab['[SEP]'] + + self.model = _WordBertModel(model_dir_or_name=model_dir_or_name, vocab=vocab, layers=layers, + pool_method=pool_method, include_cls_sep=include_cls_sep, + pooled_cls=pooled_cls, auto_truncate=auto_truncate, min_freq=2) + self.requires_grad = requires_grad - self._embed_size = len(self.model.layers)*self.model.encoder.hidden_size - + self._embed_size = len(self.model.layers) * self.model.encoder.hidden_size + def _delete_model_weights(self): del self.model - + def forward(self, words): """ 计算words的bert embedding表示。计算之前会在每句话的开始增加[CLS]在结束增加[SEP], 并根据include_cls_sep判断要不要 @@ -85,12 +101,32 @@ class BertEmbedding(ContextualEmbedding): words = self.drop_word(words) outputs = self._get_sent_reprs(words) if outputs is not None: - return self.dropout(words) + return self.dropout(outputs) outputs = self.model(words) outputs = torch.cat([*outputs], dim=-1) - + return self.dropout(outputs) + + def drop_word(self, words): + """ + 按照设定随机将words设置为unknown_index。 + :param torch.LongTensor words: batch_size x max_len + :return: + """ + if self.word_dropout > 0 and self.training: + with torch.no_grad(): + if self._word_sep_index: # 不能drop sep + sep_mask = words.eq(self._word_sep_index) + mask = torch.full_like(words, fill_value=self.word_dropout, dtype=torch.float, device=words.device) + mask = torch.bernoulli(mask).eq(1) # dropout_word越大，越多位置为1 + pad_mask = words.ne(0) + mask = pad_mask.__and__(mask) # pad的位置不为unk + words = words.masked_fill(mask, self._word_unk_index) + if self._word_sep_index: + words.masked_fill_(sep_mask, self._word_sep_index) + return words + @property def requires_grad(self): """ @@ -99,12 +135,12 @@ class BertEmbedding(ContextualEmbedding): :return: """ requires_grads = set([param.requires_grad for name, param in self.named_parameters() - if 'word_pieces_lengths' not in name]) + if 'word_pieces_lengths' not in name]) if len(requires_grads) == 1: return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for name, param in self.named_parameters(): @@ -119,27 +155,26 @@ class BertWordPieceEncoder(nn.Module): :param str model_dir_or_name: 模型所在目录或者模型的名称。默认值为 ``en-base-uncased`` :param str layers: 最终结果中的表示。以','隔开层数，可以以负数去索引倒数几层 + :param bool pooled_cls: 返回的句子开头的[CLS]是否使用预训练中的BertPool映射一下，仅在include_cls_sep时有效。如果下游任务只取 + [CLS]做预测，一般该值为True。 + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 + :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 :param bool requires_grad: 是否需要gradient。 """ - def __init__(self, model_dir_or_name: str='en-base-uncased', layers: str='-1', - requires_grad: bool=False): + + def __init__(self, model_dir_or_name: str = 'en-base-uncased', layers: str = '-1', pooled_cls: bool = False, + word_dropout=0, dropout=0, requires_grad: bool = True): super().__init__() - PRETRAIN_URL = _get_base_url('bert') - - if model_dir_or_name in PRETRAINED_BERT_MODEL_DIR: - model_name = PRETRAINED_BERT_MODEL_DIR[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(model_dir_or_name): - model_dir = model_dir_or_name - else: - raise ValueError(f"Cannot recognize {model_dir_or_name}.") - - self.model = _WordPieceBertModel(model_dir=model_dir, layers=layers) + + self.model = _WordPieceBertModel(model_dir_or_name=model_dir_or_name, layers=layers, pooled_cls=pooled_cls) + self._sep_index = self.model._sep_index + self._wordpiece_pad_index = self.model._wordpiece_pad_index + self._wordpiece_unk_index = self.model._wordpiece_unknown_index self._embed_size = len(self.model.layers) * self.model.encoder.hidden_size self.requires_grad = requires_grad - + self.word_dropout = word_dropout + self.dropout_layer = nn.Dropout(dropout) + @property def requires_grad(self): """ @@ -151,77 +186,129 @@ class BertWordPieceEncoder(nn.Module): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for name, param in self.named_parameters(): param.requires_grad = value - + @property def embed_size(self): return self._embed_size - - def index_datasets(self, *datasets, field_name): + + @property + def embedding_dim(self): + return self._embed_size + + @property + def num_embedding(self): + return self.model.encoder.config.vocab_size + + def index_datasets(self, *datasets, field_name, add_cls_sep=True): """ - 使用bert的tokenizer新生成word_pieces列加入到datasets中，并将他们设置为input。如果首尾不是 - [CLS]与[SEP]会在首尾额外加入[CLS]与[SEP], 且将word_pieces这一列的pad value设置为了bert的pad value。 + 使用bert的tokenizer新生成word_pieces列加入到datasets中，并将他们设置为input,且将word_pieces这一列的pad value设置为了 + bert的pad value。 - :param datasets: DataSet对象 - :param field_name: 基于哪一列的内容生成word_pieces列。这一列中每个数据应该是List[str]的形式。 + :param ~fastNLP.DataSet datasets: DataSet对象 + :param str field_name: 基于哪一列的内容生成word_pieces列。这一列中每个数据应该是List[str]的形式。 + :param bool add_cls_sep: 如果首尾不是[CLS]与[SEP]会在首尾额外加入[CLS]与[SEP]。 :return: """ - self.model.index_dataset(*datasets, field_name=field_name) - + self.model.index_dataset(*datasets, field_name=field_name, add_cls_sep=add_cls_sep) + def forward(self, word_pieces, token_type_ids=None): """ 计算words的bert embedding表示。传入的words中应该自行包含[CLS]与[SEP]的tag。 :param words: batch_size x max_len - :param token_type_ids: batch_size x max_len, 用于区分前一句和后一句话 + :param token_type_ids: batch_size x max_len, 用于区分前一句和后一句话. 如果不传入，则自动生成(大部分情况，都不需要输入), + 第一个[SEP]及之前为0, 第二个[SEP]及到第一个[SEP]之间为1; 第三个[SEP]及到第二个[SEP]之间为0，依次往后推。 :return: torch.FloatTensor. batch_size x max_len x (768*len(self.layers)) """ + with torch.no_grad(): + sep_mask = word_pieces.eq(self._sep_index) # batch_size x max_len + if token_type_ids is None: + sep_mask_cumsum = sep_mask.flip(dims=[-1]).cumsum(dim=-1).flip(dims=[-1]) + token_type_ids = sep_mask_cumsum.fmod(2) + if token_type_ids[0, 0].item(): # 如果开头是奇数，则需要flip一下结果，因为需要保证开头为0 + token_type_ids = token_type_ids.eq(0).long() + + word_pieces = self.drop_word(word_pieces) outputs = self.model(word_pieces, token_type_ids) outputs = torch.cat([*outputs], dim=-1) + + return self.dropout_layer(outputs) + + def drop_word(self, words): + """ + 按照设定随机将words设置为unknown_index。 - return outputs + :param torch.LongTensor words: batch_size x max_len + :return: + """ + if self.word_dropout > 0 and self.training: + with torch.no_grad(): + if self._word_sep_index: # 不能drop sep + sep_mask = words.eq(self._wordpiece_unk_index) + mask = torch.full_like(words, fill_value=self.word_dropout, dtype=torch.float, device=words.device) + mask = torch.bernoulli(mask).eq(1) # dropout_word越大，越多位置为1 + pad_mask = words.ne(self._wordpiece_pad_index) + mask = pad_mask.__and__(mask) # pad的位置不为unk + words = words.masked_fill(mask, self._word_unk_index) + if self._word_sep_index: + words.masked_fill_(sep_mask, self._wordpiece_unk_index) + return words class _WordBertModel(nn.Module): - def __init__(self, model_dir:str, vocab:Vocabulary, layers:str='-1', pool_method:str='first', include_cls_sep:bool=False): + def __init__(self, model_dir_or_name: str, vocab: Vocabulary, layers: str = '-1', pool_method: str = 'first', + include_cls_sep: bool = False, pooled_cls: bool = False, auto_truncate: bool = False, min_freq=2): super().__init__() - - self.tokenzier = BertTokenizer.from_pretrained(model_dir) - self.encoder = BertModel.from_pretrained(model_dir) + + self.tokenzier = BertTokenizer.from_pretrained(model_dir_or_name) + self.encoder = BertModel.from_pretrained(model_dir_or_name) + self._max_position_embeddings = self.encoder.config.max_position_embeddings # 检查encoder_layer_number是否合理 encoder_layer_number = len(self.encoder.encoder.layer) self.layers = list(map(int, layers.split(','))) for layer in self.layers: - if layer<0: - assert -layer<=encoder_layer_number, f"The layer index:{layer} is out of scope for " \ - f"a bert model with {encoder_layer_number} layers." + if layer < 0: + assert -layer <= encoder_layer_number, f"The layer index:{layer} is out of scope for " \ + f"a bert model with {encoder_layer_number} layers." else: - assert layer= min_freq and not vocab._is_word_no_create_entry( + word): # 出现次数大于这个次数才新增 + word_piece_dict[word] = 1 # 新增一个值 continue for word_piece in word_pieces: word_piece_dict[word_piece] = 1 @@ -242,7 +329,7 @@ class _WordBertModel(nn.Module): new_word_piece_vocab[token] = len(new_word_piece_vocab) self.tokenzier._reinit_on_new_vocab(new_word_piece_vocab) self.encoder.embeddings.word_embeddings = embed - + word_to_wordpieces = [] word_pieces_lengths = [] for word, index in vocab: @@ -254,81 +341,126 @@ class _WordBertModel(nn.Module): word_pieces = self.tokenzier.convert_tokens_to_ids(word_pieces) word_to_wordpieces.append(word_pieces) word_pieces_lengths.append(len(word_pieces)) - print("Found(Or seg into word pieces) {} words out of {}.".format(found_count, len(vocab))) self._cls_index = self.tokenzier.vocab['[CLS]'] self._sep_index = self.tokenzier.vocab['[SEP]'] - self._pad_index = vocab.padding_idx + self._word_pad_index = vocab.padding_idx self._wordpiece_pad_index = self.tokenzier.vocab['[PAD]'] # 需要用于生成word_piece + logger.info("Found(Or segment into word pieces) {} words out of {}.".format(found_count, len(vocab))) self.word_to_wordpieces = np.array(word_to_wordpieces) - self.word_pieces_lengths = nn.Parameter(torch.LongTensor(word_pieces_lengths), requires_grad=False) - print("Successfully generate word pieces.") - + self.register_buffer('word_pieces_lengths', torch.LongTensor(word_pieces_lengths)) + logger.debug("Successfully generate word pieces.") + def forward(self, words): """ :param words: torch.LongTensor, batch_size x max_len :return: num_layers x batch_size x max_len x hidden_size或者num_layers x batch_size x (max_len+2) x hidden_size """ - batch_size, max_word_len = words.size() - seq_len = words.ne(self._pad_index).sum(dim=-1) - batch_word_pieces_length = self.word_pieces_lengths[words] # batch_size x max_len - word_pieces_lengths = batch_word_pieces_length.sum(dim=-1) - max_word_piece_length = word_pieces_lengths.max().item() - # +2是由于需要加入[CLS]与[SEP] - word_pieces = words.new_full((batch_size, max_word_piece_length+2), fill_value=self._wordpiece_pad_index) - word_pieces[:, 0].fill_(self._cls_index) - batch_indexes = torch.arange(batch_size).to(words) - word_pieces[batch_indexes, word_pieces_lengths+1] = self._sep_index - attn_masks = torch.zeros_like(word_pieces) - # 1. 获取words的word_pieces的id，以及对应的span范围 - word_indexes = words.tolist() - for i in range(batch_size): - word_pieces_i = list(chain(*self.word_to_wordpieces[word_indexes[i]])) - word_pieces[i, 1:len(word_pieces_i)+1] = torch.LongTensor(word_pieces_i) - attn_masks[i, :len(word_pieces_i)+2].fill_(1) - # TODO 截掉长度超过的部分。 + with torch.no_grad(): + batch_size, max_word_len = words.size() + word_mask = words.ne(self._word_pad_index) # 为1的地方有word + seq_len = word_mask.sum(dim=-1) + batch_word_pieces_length = self.word_pieces_lengths[words].masked_fill(word_mask.eq(0), + 0) # batch_size x max_len + word_pieces_lengths = batch_word_pieces_length.sum(dim=-1) # batch_size + word_piece_length = batch_word_pieces_length.sum(dim=-1).max().item() # 表示word piece的长度(包括padding) + if word_piece_length + 2 > self._max_position_embeddings: + if self.auto_truncate: + word_pieces_lengths = word_pieces_lengths.masked_fill( + word_pieces_lengths + 2 > self._max_position_embeddings, + self._max_position_embeddings - 2) + else: + raise RuntimeError( + "After split words into word pieces, the lengths of word pieces are longer than the " + f"maximum allowed sequence length:{self._max_position_embeddings} of bert. You can set " + f"`auto_truncate=True` for BertEmbedding to automatically truncate overlong input.") + + # +2是由于需要加入[CLS]与[SEP] + word_pieces = words.new_full((batch_size, min(word_piece_length + 2, self._max_position_embeddings)), + fill_value=self._wordpiece_pad_index) + attn_masks = torch.zeros_like(word_pieces) + # 1. 获取words的word_pieces的id，以及对应的span范围 + word_indexes = words.cpu().numpy() + for i in range(batch_size): + word_pieces_i = list(chain(*self.word_to_wordpieces[word_indexes[i, :seq_len[i]]])) + if self.auto_truncate and len(word_pieces_i) > self._max_position_embeddings - 2: + word_pieces_i = word_pieces_i[:self._max_position_embeddings - 2] + word_pieces[i, 1:word_pieces_lengths[i] + 1] = torch.LongTensor(word_pieces_i) + attn_masks[i, :word_pieces_lengths[i] + 2].fill_(1) + # 添加[cls]和[sep] + word_pieces[:, 0].fill_(self._cls_index) + batch_indexes = torch.arange(batch_size).to(words) + word_pieces[batch_indexes, word_pieces_lengths + 1] = self._sep_index + if self._has_sep_in_vocab: # 但[SEP]在vocab中出现应该才会需要token_ids + sep_mask = word_pieces.eq(self._sep_index).long() # batch_size x max_len + sep_mask_cumsum = sep_mask.flip(dims=[-1]).cumsum(dim=-1).flip(dims=[-1]) + token_type_ids = sep_mask_cumsum.fmod(2) + if token_type_ids[0, 0].item(): # 如果开头是奇数，则需要flip一下结果，因为需要保证开头为0 + token_type_ids = token_type_ids.eq(0).long() + else: + token_type_ids = torch.zeros_like(word_pieces) # 2. 获取hidden的结果，根据word_pieces进行对应的pool计算 # all_outputs: [batch_size x max_len x hidden_size, batch_size x max_len x hidden_size, ...] - bert_outputs, _ = self.encoder(word_pieces, token_type_ids=None, attention_mask=attn_masks, - output_all_encoded_layers=True) - # output_layers = [self.layers] # len(self.layers) x batch_size x max_word_piece_length x hidden_size - + bert_outputs, pooled_cls = self.encoder(word_pieces, token_type_ids=token_type_ids, attention_mask=attn_masks, + output_all_encoded_layers=True) + # output_layers = [self.layers] # len(self.layers) x batch_size x real_word_piece_length x hidden_size + if self.include_cls_sep: - outputs = bert_outputs[-1].new_zeros(len(self.layers), batch_size, max_word_len + 2, - bert_outputs[-1].size(-1)) s_shift = 1 + outputs = bert_outputs[-1].new_zeros(len(self.layers), batch_size, max_word_len + 2, + bert_outputs[-1].size(-1)) + else: + s_shift = 0 outputs = bert_outputs[-1].new_zeros(len(self.layers), batch_size, max_word_len, bert_outputs[-1].size(-1)) - s_shift = 0 batch_word_pieces_cum_length = batch_word_pieces_length.new_zeros(batch_size, max_word_len + 1) batch_word_pieces_cum_length[:, 1:] = batch_word_pieces_length.cumsum(dim=-1) # batch_size x max_len + + if self.pool_method == 'first': + batch_word_pieces_cum_length = batch_word_pieces_cum_length[:, :seq_len.max()] + batch_word_pieces_cum_length.masked_fill_(batch_word_pieces_cum_length.ge(word_piece_length), 0) + _batch_indexes = batch_indexes[:, None].expand((batch_size, batch_word_pieces_cum_length.size(1))) + elif self.pool_method == 'last': + batch_word_pieces_cum_length = batch_word_pieces_cum_length[:, 1:seq_len.max()+1] - 1 + batch_word_pieces_cum_length.masked_fill_(batch_word_pieces_cum_length.ge(word_piece_length), 0) + _batch_indexes = batch_indexes[:, None].expand((batch_size, batch_word_pieces_cum_length.size(1))) + for l_index, l in enumerate(self.layers): output_layer = bert_outputs[l] + real_word_piece_length = output_layer.size(1) - 2 + if word_piece_length > real_word_piece_length: # 如果实际上是截取出来的 + paddings = output_layer.new_zeros(batch_size, + word_piece_length - real_word_piece_length, + output_layer.size(2)) + output_layer = torch.cat((output_layer, paddings), dim=1).contiguous() # 从word_piece collapse到word的表示 truncate_output_layer = output_layer[:, 1:-1] # 删除[CLS]与[SEP] batch_size x len x hidden_size - outputs_seq_len = seq_len + s_shift if self.pool_method == 'first': - for i in range(batch_size): - i_word_pieces_cum_length = batch_word_pieces_cum_length[i, :seq_len[i]] # 每个word的start位置 - outputs[l_index, i, s_shift:outputs_seq_len[i]] = truncate_output_layer[i, i_word_pieces_cum_length] # num_layer x batch_size x len x hidden_size + tmp = truncate_output_layer[_batch_indexes, batch_word_pieces_cum_length] + tmp = tmp.masked_fill(word_mask[:, :batch_word_pieces_cum_length.size(1), None].eq(0), 0) + outputs[l_index, :, s_shift:batch_word_pieces_cum_length.size(1)+s_shift] = tmp + elif self.pool_method == 'last': - for i in range(batch_size): - i_word_pieces_cum_length = batch_word_pieces_cum_length[i, 1:seq_len[i]+1] - 1 # 每个word的end - outputs[l_index, i, s_shift:outputs_seq_len[i]] = truncate_output_layer[i, i_word_pieces_cum_length] + tmp = truncate_output_layer[_batch_indexes, batch_word_pieces_cum_length] + tmp = tmp.masked_fill(word_mask[:, :batch_word_pieces_cum_length.size(1), None].eq(0), 0) + outputs[l_index, :, s_shift:batch_word_pieces_cum_length.size(1)+s_shift] = tmp elif self.pool_method == 'max': for i in range(batch_size): for j in range(seq_len[i]): - start, end = batch_word_pieces_cum_length[i, j], batch_word_pieces_cum_length[i, j+1] - outputs[l_index, i, j+s_shift], _ = torch.max(truncate_output_layer[i, start:end], dim=-2) + start, end = batch_word_pieces_cum_length[i, j], batch_word_pieces_cum_length[i, j + 1] + outputs[l_index, i, j + s_shift], _ = torch.max(truncate_output_layer[i, start:end], dim=-2) else: for i in range(batch_size): for j in range(seq_len[i]): - start, end = batch_word_pieces_cum_length[i, j], batch_word_pieces_cum_length[i, j+1] - outputs[l_index, i, j+s_shift] = torch.mean(truncate_output_layer[i, start:end], dim=-2) + start, end = batch_word_pieces_cum_length[i, j], batch_word_pieces_cum_length[i, j + 1] + outputs[l_index, i, j + s_shift] = torch.mean(truncate_output_layer[i, start:end], dim=-2) if self.include_cls_sep: - outputs[l_index, :, 0] = output_layer[:, 0] - outputs[l_index, batch_indexes, seq_len+s_shift] = output_layer[batch_indexes, seq_len+s_shift] + if l in (len(bert_outputs) - 1, -1) and self.pooled_cls: + outputs[l_index, :, 0] = pooled_cls + else: + outputs[l_index, :, 0] = output_layer[:, 0] + outputs[l_index, batch_indexes, seq_len + s_shift] = output_layer[batch_indexes, seq_len + s_shift] + # 3. 最终的embedding结果 return outputs - diff --git a/fastNLP/embeddings/char_embedding.py b/fastNLP/embeddings/char_embedding.py index b9e6659e..59109206 100644 --- a/fastNLP/embeddings/char_embedding.py +++ b/fastNLP/embeddings/char_embedding.py @@ -3,27 +3,35 @@ 词的index而不需要使用词语中的char的index来获取表达。 """ +__all__ = [ + "CNNCharEmbedding", + "LSTMCharEmbedding" +] import torch import torch.nn as nn import torch.nn.functional as F from typing import List +from .static_embedding import StaticEmbedding from ..modules.encoder.lstm import LSTM from ..core.vocabulary import Vocabulary from .embedding import TokenEmbedding from .utils import _construct_char_vocab_from_vocab +from .utils import get_embeddings +from ..core import logger class CNNCharEmbedding(TokenEmbedding): """ - 别名：:class:`fastNLP.embeddings.CNNCharEmbedding` :class:`fastNLP.embeddings.char_embedding.CNNCharEmbedding` - 使用CNN生成character embedding。CNN的结构为, embed(x) -> Dropout(x) -> CNN(x) -> activation(x) -> pool -> fc -> Dropout. 不同的kernel大小的fitler结果是concat起来然后通过一层fully connected layer, 然后输出word的表示。 Example:: + >>> import torch + >>> from fastNLP import Vocabulary + >>> from fastNLP.embeddings import CNNCharEmbedding >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) >>> embed = CNNCharEmbedding(vocab, embed_size=50) >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) @@ -32,8 +40,8 @@ class CNNCharEmbedding(TokenEmbedding): >>> # torch.Size([1, 5，50]) :param vocab: 词表 - :param embed_size: 该word embedding的大小，默认值为50. - :param char_emb_size: character的embed的大小。character是从vocab中生成的。默认值为50. + :param embed_size: 该CNNCharEmbedding的输出维度大小，默认值为50. + :param char_emb_size: character的embed的维度。character是从vocab中生成的。默认值为50. :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 :param float dropout: 以多大的概率drop分布式表示与char embedding的输出。 :param filter_nums: filter的数量. 长度需要和kernels一致。默认值为[40, 30, 20]. @@ -41,17 +49,20 @@ class CNNCharEmbedding(TokenEmbedding): :param pool_method: character的表示在合成一个表示时所使用的pool方法，支持'avg', 'max'. :param activation: CNN之后使用的激活方法，支持'relu', 'sigmoid', 'tanh' 或者自定义函数. :param min_char_freq: character的最少出现次数。默认值为2. + :param pre_train_char_embed: 可以有两种方式调用预训练好的character embedding：第一种是传入embedding文件夹 + (文件夹下应该只有一个以.txt作为后缀的文件)或文件路径；第二种是传入embedding的名称，第二种情况将自动查看缓存中是否存在该模型， + 没有的话将自动下载。如果输入为None则使用embedding_dim的维度随机初始化一个embedding. """ - def __init__(self, vocab: Vocabulary, embed_size: int=50, char_emb_size: int=50, word_dropout:float=0, - dropout:float=0.5, filter_nums: List[int]=(40, 30, 20), kernel_sizes: List[int]=(5, 3, 1), - pool_method: str='max', activation='relu', min_char_freq: int=2): + + def __init__(self, vocab: Vocabulary, embed_size: int = 50, char_emb_size: int = 50, word_dropout: float = 0, + dropout: float = 0, filter_nums: List[int] = (40, 30, 20), kernel_sizes: List[int] = (5, 3, 1), + pool_method: str = 'max', activation='relu', min_char_freq: int = 2, pre_train_char_embed: str = None): super(CNNCharEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - + for kernel in kernel_sizes: assert kernel % 2 == 1, "Only odd kernel is allowed." - + assert pool_method in ('max', 'avg') - self.dropout = nn.Dropout(dropout) self.pool_method = pool_method # activation function if isinstance(activation, str): @@ -68,32 +79,35 @@ class CNNCharEmbedding(TokenEmbedding): else: raise Exception( "Undefined activation function: choose from: [relu, tanh, sigmoid, or a callable function]") - - print("Start constructing character vocabulary.") + + logger.info("Start constructing character vocabulary.") # 建立char的词表 self.char_vocab = _construct_char_vocab_from_vocab(vocab, min_freq=min_char_freq) self.char_pad_index = self.char_vocab.padding_idx - print(f"In total, there are {len(self.char_vocab)} distinct characters.") + logger.info(f"In total, there are {len(self.char_vocab)} distinct characters.") # 对vocab进行index max_word_len = max(map(lambda x: len(x[0]), vocab)) - self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab), max_word_len), - fill_value=self.char_pad_index, dtype=torch.long), - requires_grad=False) - self.word_lengths = nn.Parameter(torch.zeros(len(vocab)).long(), requires_grad=False) + self.register_buffer('words_to_chars_embedding', torch.full((len(vocab), max_word_len), + fill_value=self.char_pad_index, dtype=torch.long)) + self.register_buffer('word_lengths', torch.zeros(len(vocab)).long()) for word, index in vocab: # if index!=vocab.padding_idx: # 如果是pad的话，直接就为pad_value了。修改为不区分pad, 这样所有的也是同一个embed self.words_to_chars_embedding[index, :len(word)] = \ torch.LongTensor([self.char_vocab.to_index(c) for c in word]) self.word_lengths[index] = len(word) - self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) - + # self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) + if pre_train_char_embed: + self.char_embedding = StaticEmbedding(self.char_vocab, model_dir_or_name=pre_train_char_embed) + else: + self.char_embedding = get_embeddings((len(self.char_vocab), char_emb_size)) + self.convs = nn.ModuleList([nn.Conv1d( char_emb_size, filter_nums[i], kernel_size=kernel_sizes[i], bias=True, padding=kernel_sizes[i] // 2) for i in range(len(kernel_sizes))]) self._embed_size = embed_size self.fc = nn.Linear(sum(filter_nums), embed_size) - self.init_param() - + self.reset_parameters() + def forward(self, words): """ 输入words的index后，生成对应的words的表示。 @@ -104,14 +118,14 @@ class CNNCharEmbedding(TokenEmbedding): words = self.drop_word(words) batch_size, max_len = words.size() chars = self.words_to_chars_embedding[words] # batch_size x max_len x max_word_len - word_lengths = self.word_lengths[words] # batch_size x max_len + word_lengths = self.word_lengths[words] # batch_size x max_len max_word_len = word_lengths.max() chars = chars[:, :, :max_word_len] # 为1的地方为mask chars_masks = chars.eq(self.char_pad_index) # batch_size x max_len x max_word_len 如果为0, 说明是padding的位置了 chars = self.char_embedding(chars) # batch_size x max_len x max_word_len x embed_size chars = self.dropout(chars) - reshaped_chars = chars.reshape(batch_size*max_len, max_word_len, -1) + reshaped_chars = chars.reshape(batch_size * max_len, max_word_len, -1) reshaped_chars = reshaped_chars.transpose(1, 2) # B' x E x M conv_chars = [conv(reshaped_chars).transpose(1, 2).reshape(batch_size, max_len, max_word_len, -1) for conv in self.convs] @@ -119,13 +133,13 @@ class CNNCharEmbedding(TokenEmbedding): conv_chars = self.activation(conv_chars) if self.pool_method == 'max': conv_chars = conv_chars.masked_fill(chars_masks.unsqueeze(-1), float('-inf')) - chars, _ = torch.max(conv_chars, dim=-2) # batch_size x max_len x sum(filters) + chars, _ = torch.max(conv_chars, dim=-2) # batch_size x max_len x sum(filters) else: conv_chars = conv_chars.masked_fill(chars_masks.unsqueeze(-1), 0) - chars = torch.sum(conv_chars, dim=-2)/chars_masks.eq(0).sum(dim=-1, keepdim=True).float() + chars = torch.sum(conv_chars, dim=-2) / chars_masks.eq(0).sum(dim=-1, keepdim=True).float() chars = self.fc(chars) return self.dropout(chars) - + @property def requires_grad(self): """ @@ -141,19 +155,21 @@ class CNNCharEmbedding(TokenEmbedding): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for name, param in self.named_parameters(): if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能加入到requires_grad中 continue param.requires_grad = value - - def init_param(self): + + def reset_parameters(self): for name, param in self.named_parameters(): if 'words_to_chars_embedding' in name or 'word_lengths' in name: # 这个不能reset continue - if param.data.dim()>1: + if 'char_embedding' in name: + continue + if param.data.dim() > 1: nn.init.xavier_uniform_(param, 1) else: nn.init.uniform_(param, -1, 1) @@ -161,12 +177,13 @@ class CNNCharEmbedding(TokenEmbedding): class LSTMCharEmbedding(TokenEmbedding): """ - 别名：:class:`fastNLP.embeddings.LSTMCharEmbedding` :class:`fastNLP.embeddings.char_embedding.LSTMCharEmbedding` - 使用LSTM的方式对character进行encode. embed(x) -> Dropout(x) -> LSTM(x) -> activation(x) -> pool -> Dropout Example:: + >>> import torch + >>> from fastNLP import Vocabulary + >>> from fastNLP.embeddings import LSTMCharEmbedding >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) >>> embed = LSTMCharEmbedding(vocab, embed_size=50) >>> words = torch.LongTensor([[vocab.to_index(word) for word in "The whether is good .".split()]]) @@ -175,8 +192,8 @@ class LSTMCharEmbedding(TokenEmbedding): >>> # torch.Size([1, 5，50]) :param vocab: 词表 - :param embed_size: embedding的大小。默认值为50. - :param char_emb_size: character的embedding的大小。默认值为50. + :param embed_size: LSTMCharEmbedding的输出维度。默认值为50. + :param char_emb_size: character的embedding的维度。默认值为50. :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 :param dropout: 以多大概率drop character embedding的输出以及最终的word的输出。 :param hidden_size: LSTM的中间hidden的大小，如果为bidirectional的，hidden会除二，默认为50. @@ -184,17 +201,21 @@ class LSTMCharEmbedding(TokenEmbedding): :param activation: 激活函数，支持'relu', 'sigmoid', 'tanh', 或者自定义函数. :param min_char_freq: character的最小出现次数。默认值为2. :param bidirectional: 是否使用双向的LSTM进行encode。默认值为True。 + :param pre_train_char_embed: 可以有两种方式调用预训练好的character embedding：第一种是传入embedding文件夹 + (文件夹下应该只有一个以.txt作为后缀的文件)或文件路径；第二种是传入embedding的名称，第二种情况将自动查看缓存中是否存在该模型， + 没有的话将自动下载。如果输入为None则使用embedding_dim的维度随机初始化一个embedding. """ - def __init__(self, vocab: Vocabulary, embed_size: int=50, char_emb_size: int=50, word_dropout:float=0, - dropout:float=0.5, hidden_size=50,pool_method: str='max', activation='relu', min_char_freq: int=2, - bidirectional=True): - super(LSTMCharEmbedding, self).__init__(vocab) - + + def __init__(self, vocab: Vocabulary, embed_size: int = 50, char_emb_size: int = 50, word_dropout: float = 0, + dropout: float = 0, hidden_size=50, pool_method: str = 'max', activation='relu', + min_char_freq: int = 2, + bidirectional=True, pre_train_char_embed: str = None): + super(LSTMCharEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) + assert hidden_size % 2 == 0, "Only even kernel is allowed." - + assert pool_method in ('max', 'avg') self.pool_method = pool_method - self.dropout = nn.Dropout(dropout) # activation function if isinstance(activation, str): if activation.lower() == 'relu': @@ -210,32 +231,35 @@ class LSTMCharEmbedding(TokenEmbedding): else: raise Exception( "Undefined activation function: choose from: [relu, tanh, sigmoid, or a callable function]") - - print("Start constructing character vocabulary.") + + logger.info("Start constructing character vocabulary.") # 建立char的词表 self.char_vocab = _construct_char_vocab_from_vocab(vocab, min_freq=min_char_freq) self.char_pad_index = self.char_vocab.padding_idx - print(f"In total, there are {len(self.char_vocab)} distinct characters.") + logger.info(f"In total, there are {len(self.char_vocab)} distinct characters.") # 对vocab进行index self.max_word_len = max(map(lambda x: len(x[0]), vocab)) - self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab), self.max_word_len), - fill_value=self.char_pad_index, dtype=torch.long), - requires_grad=False) - self.word_lengths = nn.Parameter(torch.zeros(len(vocab)).long(), requires_grad=False) + self.register_buffer('words_to_chars_embedding', torch.full((len(vocab), self.max_word_len), + fill_value=self.char_pad_index, dtype=torch.long)) + self.register_buffer('word_lengths', torch.zeros(len(vocab)).long()) for word, index in vocab: # if index!=vocab.padding_idx: # 如果是pad的话，直接就为pad_value了. 修改为不区分pad与否 self.words_to_chars_embedding[index, :len(word)] = \ torch.LongTensor([self.char_vocab.to_index(c) for c in word]) self.word_lengths[index] = len(word) - self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) - + # self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) + if pre_train_char_embed: + self.char_embedding = StaticEmbedding(self.char_vocab, pre_train_char_embed) + else: + self.char_embedding = nn.Embedding(len(self.char_vocab), char_emb_size) + self.fc = nn.Linear(hidden_size, embed_size) hidden_size = hidden_size // 2 if bidirectional else hidden_size - + self.lstm = LSTM(char_emb_size, hidden_size, bidirectional=bidirectional, batch_first=True) self._embed_size = embed_size self.bidirectional = bidirectional - + def forward(self, words): """ 输入words的index后，生成对应的words的表示。 @@ -257,7 +281,7 @@ class LSTMCharEmbedding(TokenEmbedding): char_seq_len = chars_masks.eq(0).sum(dim=-1).reshape(batch_size * max_len) lstm_chars = self.lstm(reshaped_chars, char_seq_len)[0].reshape(batch_size, max_len, max_word_len, -1) # B x M x M x H - + lstm_chars = self.activation(lstm_chars) if self.pool_method == 'max': lstm_chars = lstm_chars.masked_fill(chars_masks.unsqueeze(-1), float('-inf')) @@ -265,11 +289,11 @@ class LSTMCharEmbedding(TokenEmbedding): else: lstm_chars = lstm_chars.masked_fill(chars_masks.unsqueeze(-1), 0) chars = torch.sum(lstm_chars, dim=-2) / chars_masks.eq(0).sum(dim=-1, keepdim=True).float() - + chars = self.fc(chars) - + return self.dropout(chars) - + @property def requires_grad(self): """ @@ -286,7 +310,7 @@ class LSTMCharEmbedding(TokenEmbedding): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for name, param in self.named_parameters(): diff --git a/fastNLP/embeddings/contextual_embedding.py b/fastNLP/embeddings/contextual_embedding.py index 1831af4e..9910a44b 100644 --- a/fastNLP/embeddings/contextual_embedding.py +++ b/fastNLP/embeddings/contextual_embedding.py @@ -1,20 +1,30 @@ +""" +.. todo:: + doc +""" + +__all__ = [ + "ContextualEmbedding" +] from abc import abstractmethod + import torch -from ..core.vocabulary import Vocabulary -from ..core.dataset import DataSet +from .embedding import TokenEmbedding +from ..core import logger from ..core.batch import DataSetIter +from ..core.dataset import DataSet from ..core.sampler import SequentialSampler from ..core.utils import _move_model_to_device, _get_model_device -from .embedding import TokenEmbedding +from ..core.vocabulary import Vocabulary class ContextualEmbedding(TokenEmbedding): - def __init__(self, vocab: Vocabulary, word_dropout:float=0.0, dropout:float=0.0): + def __init__(self, vocab: Vocabulary, word_dropout: float = 0.0, dropout: float = 0.0): super(ContextualEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - - def add_sentence_cache(self, *datasets, batch_size=32, device='cpu', delete_weights: bool=True): + + def add_sentence_cache(self, *datasets, batch_size=32, device='cpu', delete_weights: bool = True): """ 由于动态embedding生成比较耗时，所以可以把每句话embedding缓存下来，这样就不需要每次都运行生成过程。 @@ -29,14 +39,14 @@ class ContextualEmbedding(TokenEmbedding): assert isinstance(dataset, DataSet), "Only fastNLP.DataSet object is allowed." assert 'words' in dataset.get_input_name(), "`words` field has to be set as input." except Exception as e: - print(f"Exception happens at {index} dataset.") + logger.error(f"Exception happens at {index} dataset.") raise e - + sent_embeds = {} _move_model_to_device(self, device=device) device = _get_model_device(self) pad_index = self._word_vocab.padding_idx - print("Start to calculate sentence representations.") + logger.info("Start to calculate sentence representations.") with torch.no_grad(): for index, dataset in enumerate(datasets): try: @@ -51,18 +61,18 @@ class ContextualEmbedding(TokenEmbedding): word_embeds = self(words).detach().cpu().numpy() for b in range(words.size(0)): length = seq_len_from_behind[b] - if length==0: + if length == 0: sent_embeds[tuple(words_list[b][:seq_len[b]])] = word_embeds[b] else: sent_embeds[tuple(words_list[b][:seq_len[b]])] = word_embeds[b, :-length] except Exception as e: - print(f"Exception happens at {index} dataset.") + logger.error(f"Exception happens at {index} dataset.") raise e - print("Finish calculating sentence representations.") + logger.info("Finish calculating sentence representations.") self.sent_embeds = sent_embeds if delete_weights: self._delete_model_weights() - + def _get_sent_reprs(self, words): """ 获取sentence的表示，如果有缓存，则返回缓存的值; 没有缓存则返回None @@ -85,12 +95,12 @@ class ContextualEmbedding(TokenEmbedding): embeds[i, :len(embed)] = torch.FloatTensor(embed).to(words.device) return embeds return None - + @abstractmethod def _delete_model_weights(self): """删除计算表示的模型以节省资源""" raise NotImplementedError - + def remove_sentence_cache(self): """ 删除缓存的句子表示. 删除之后如果模型权重没有被删除，将开始使用动态计算权重。 diff --git a/fastNLP/embeddings/elmo_embedding.py b/fastNLP/embeddings/elmo_embedding.py index af94e8ec..d19a3577 100644 --- a/fastNLP/embeddings/elmo_embedding.py +++ b/fastNLP/embeddings/elmo_embedding.py @@ -1,6 +1,13 @@ +""" +.. todo:: + doc +""" -import os +__all__ = [ + "ElmoEmbedding" +] +import os import torch import torch.nn as nn import torch.nn.functional as F @@ -8,19 +15,20 @@ import json import codecs from ..core.vocabulary import Vocabulary -from ..io.file_utils import cached_path, _get_base_url, PRETRAINED_ELMO_MODEL_DIR +from ..io.file_utils import cached_path, _get_embedding_url, PRETRAINED_ELMO_MODEL_DIR from ..modules.encoder._elmo import ElmobiLm, ConvTokenEmbedder from .contextual_embedding import ContextualEmbedding - +from ..core import logger class ElmoEmbedding(ContextualEmbedding): """ - 别名：:class:`fastNLP.embeddings.ElmoEmbedding` :class:`fastNLP.embeddings.elmo_embedding.ElmoEmbedding` - 使用ELMo的embedding。初始化之后，只需要传入words就可以得到对应的embedding。当前支持的使用名称初始化的模型有以下的这些(待补充) Example:: + >>> import torch + >>> from fastNLP import Vocabulary + >>> from fastNLP.embeddings import ElmoEmbedding >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) >>> # 使用不同层的concat的结果 >>> embed = ElmoEmbedding(vocab, model_dir_or_name='en', layers='1,2', requires_grad=False) @@ -37,7 +45,7 @@ class ElmoEmbedding(ContextualEmbedding): :param model_dir_or_name: 可以有两种方式调用预训练好的ELMo embedding：第一种是传入ELMo所在文件夹，该文件夹下面应该有两个文件， 其中一个是以json为后缀的配置文件，另一个是以pkl为后缀的权重文件；第二种是传入ELMo版本的名称，将自动查看缓存中是否存在该模型， 没有的话将自动下载并缓存。 - :param layers: str, 指定返回的层数, 以,隔开不同的层。如果要返回第二层的结果'2', 返回后两层的结果'1,2'。不同的层的结果 + :param layers: str, 指定返回的层数(从0开始), 以,隔开不同的层。如果要返回第二层的结果'2', 返回后两层的结果'1,2'。不同的层的结果 按照这个顺序concat起来，默认为'2'。'mix'会使用可学习的权重结合不同层的表示(权重是否可训练与requires_grad保持一致， 初始化权重对三层结果进行mean-pooling, 可以通过ElmoEmbedding.set_mix_weights_requires_grad()方法只将mix weights设置为可学习。) :param requires_grad: bool, 该层是否需要gradient, 默认为False. @@ -46,24 +54,23 @@ class ElmoEmbedding(ContextualEmbedding): :param cache_word_reprs: 可以选择对word的表示进行cache; 设置为True的话，将在初始化的时候为每个word生成对应的embedding， 并删除character encoder，之后将直接使用cache的embedding。默认为False。 """ - + def __init__(self, vocab: Vocabulary, model_dir_or_name: str = 'en', layers: str = '2', requires_grad: bool = False, word_dropout=0.0, dropout=0.0, cache_word_reprs: bool = False): super(ElmoEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - + # 根据model_dir_or_name检查是否存在并下载 if model_dir_or_name.lower() in PRETRAINED_ELMO_MODEL_DIR: - PRETRAIN_URL = _get_base_url('elmo') - model_name = PRETRAINED_ELMO_MODEL_DIR[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) + model_url = _get_embedding_url('elmo', model_dir_or_name.lower()) + model_dir = cached_path(model_url, name='embedding') # 检查是否存在 - elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): + elif os.path.isdir(os.path.abspath(os.path.expanduser(model_dir_or_name))): model_dir = model_dir_or_name else: raise ValueError(f"Cannot recognize {model_dir_or_name}.") self.model = _ElmoModel(model_dir, vocab, cache_word_reprs=cache_word_reprs) - + num_layers = self.model.encoder.num_layers + if layers == 'mix': self.layer_weights = nn.Parameter(torch.zeros(self.model.config['lstm']['n_layers'] + 1), requires_grad=requires_grad) @@ -72,22 +79,22 @@ class ElmoEmbedding(ContextualEmbedding): self._embed_size = self.model.config['lstm']['projection_dim'] * 2 else: layers = list(map(int, layers.split(','))) - assert len(layers) > 0, "Must choose one output" + assert len(layers) > 0, "Must choose at least one output, but got None." for layer in layers: - assert 0 <= layer <= 2, "Layer index should be in range [0, 2]." + assert 0 <= layer <= num_layers, f"Layer index should be in range [0, {num_layers}], but got {layer}." self.layers = layers self._get_outputs = self._get_layer_outputs self._embed_size = len(self.layers) * self.model.config['lstm']['projection_dim'] * 2 - + self.requires_grad = requires_grad - + def _get_mixed_outputs(self, outputs): # outputs: num_layers x batch_size x max_len x hidden_size # return: batch_size x max_len x hidden_size weights = F.softmax(self.layer_weights + 1 / len(outputs), dim=0).to(outputs) outputs = torch.einsum('l,lbij->bij', weights, outputs) return self.gamma.to(outputs) * outputs - + def set_mix_weights_requires_grad(self, flag=True): """ 当初始化ElmoEmbedding时layers被设置为mix时，可以通过调用该方法设置mix weights是否可训练。如果layers不是mix，调用 @@ -99,15 +106,15 @@ class ElmoEmbedding(ContextualEmbedding): if hasattr(self, 'layer_weights'): self.layer_weights.requires_grad = flag self.gamma.requires_grad = flag - + def _get_layer_outputs(self, outputs): if len(self.layers) == 1: outputs = outputs[self.layers[0]] else: outputs = torch.cat(tuple([*outputs[self.layers]]), dim=-1) - + return outputs - + def forward(self, words: torch.LongTensor): """ 计算words的elmo embedding表示。根据elmo文章中介绍的ELMO实际上是有2L+1层结果，但是为了让结果比较容易拆分，token的 @@ -124,12 +131,12 @@ class ElmoEmbedding(ContextualEmbedding): outputs = self.model(words) outputs = self._get_outputs(outputs) return self.dropout(outputs) - + def _delete_model_weights(self): for name in ['layers', 'model', 'layer_weights', 'gamma']: if hasattr(self, name): delattr(self, name) - + @property def requires_grad(self): """ @@ -143,7 +150,7 @@ class ElmoEmbedding(ContextualEmbedding): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for name, param in self.named_parameters(): @@ -161,7 +168,7 @@ class _ElmoModel(nn.Module): (4) 设计一个保存token的embedding，允许缓存word的表示。 """ - + def __init__(self, model_dir: str, vocab: Vocabulary = None, cache_word_reprs: bool = False): super(_ElmoModel, self).__init__() self.model_dir = model_dir @@ -182,18 +189,18 @@ class _ElmoModel(nn.Module): raise Exception(f"Multiple config files(*.json) or weight files(*.hdf5) detected in {model_dir}.") elif config_count == 0 or weight_count == 0: raise Exception(f"No config file or weight file found in {model_dir}") - - config = json.load(open(os.path.join(model_dir, config_file), 'r')) + with open(os.path.join(model_dir, config_file), 'r') as config_f: + config = json.load(config_f) self.weight_file = os.path.join(model_dir, weight_file) self.config = config - + OOV_TAG = '' PAD_TAG = '' BOS_TAG = '' EOS_TAG = '' BOW_TAG = '' EOW_TAG = '' - + # For the model trained with character-based word encoder. char_lexicon = {} with codecs.open(os.path.join(model_dir, 'char.dic'), 'r', encoding='utf-8') as fpi: @@ -203,29 +210,29 @@ class _ElmoModel(nn.Module): tokens.insert(0, '\u3000') token, i = tokens char_lexicon[token] = int(i) - + # 做一些sanity check for special_word in [PAD_TAG, OOV_TAG, BOW_TAG, EOW_TAG]: assert special_word in char_lexicon, f"{special_word} not found in char.dic." - + # 从vocab中构建char_vocab char_vocab = Vocabulary(unknown=OOV_TAG, padding=PAD_TAG) # 需要保证与在里面 char_vocab.add_word_lst([BOW_TAG, EOW_TAG, BOS_TAG, EOS_TAG]) - + for word, index in vocab: char_vocab.add_word_lst(list(word)) - + self.bos_index, self.eos_index, self._pad_index = len(vocab), len(vocab) + 1, vocab.padding_idx # 根据char_lexicon调整, 多设置一位，是预留给word padding的(该位置的char表示为全0表示) char_emb_layer = nn.Embedding(len(char_vocab) + 1, int(config['char_cnn']['embedding']['dim']), padding_idx=len(char_vocab)) - + # 读入预训练权重 这里的elmo_model 包含char_cnn和 lstm 的 state_dict elmo_model = torch.load(os.path.join(self.model_dir, weight_file), map_location='cpu') - + char_embed_weights = elmo_model["char_cnn"]['char_emb_layer.weight'] - + found_char_count = 0 for char, index in char_vocab: # 调整character embedding if char in char_lexicon: @@ -234,15 +241,13 @@ class _ElmoModel(nn.Module): else: index_in_pre = char_lexicon[OOV_TAG] char_emb_layer.weight.data[index] = char_embed_weights[index_in_pre] - - print(f"{found_char_count} out of {len(char_vocab)} characters were found in pretrained elmo embedding.") + + logger.info(f"{found_char_count} out of {len(char_vocab)} characters were found in pretrained elmo embedding.") # 生成words到chars的映射 max_chars = config['char_cnn']['max_characters_per_token'] - - self.words_to_chars_embedding = nn.Parameter(torch.full((len(vocab) + 2, max_chars), + self.register_buffer('words_to_chars_embedding', torch.full((len(vocab) + 2, max_chars), fill_value=len(char_vocab), - dtype=torch.long), - requires_grad=False) + dtype=torch.long)) for word, index in list(iter(vocab)) + [(BOS_TAG, len(vocab)), (EOS_TAG, len(vocab) + 1)]: if len(word) + 2 > max_chars: word = word[:max_chars - 2] @@ -257,29 +262,29 @@ class _ElmoModel(nn.Module): char_vocab.to_index(EOW_TAG)] char_ids += [char_vocab.to_index(PAD_TAG)] * (max_chars - len(char_ids)) self.words_to_chars_embedding[index] = torch.LongTensor(char_ids) - + self.char_vocab = char_vocab - + self.token_embedder = ConvTokenEmbedder( config, self.weight_file, None, char_emb_layer) elmo_model["char_cnn"]['char_emb_layer.weight'] = char_emb_layer.weight self.token_embedder.load_state_dict(elmo_model["char_cnn"]) - + self.output_dim = config['lstm']['projection_dim'] - + # lstm encoder self.encoder = ElmobiLm(config) self.encoder.load_state_dict(elmo_model["lstm"]) - + if cache_word_reprs: if config['char_cnn']['embedding']['dim'] > 0: # 只有在使用了chars的情况下有用 - print("Start to generate cache word representations.") + logger.info("Start to generate cache word representations.") batch_size = 320 # bos eos word_size = self.words_to_chars_embedding.size(0) num_batches = word_size // batch_size + \ int(word_size % batch_size != 0) - + self.cached_word_embedding = nn.Embedding(word_size, config['lstm']['projection_dim']) with torch.no_grad(): @@ -290,12 +295,12 @@ class _ElmoModel(nn.Module): word_reprs = self.token_embedder(words.unsqueeze(1), chars).detach() # batch_size x 1 x config['encoder']['projection_dim'] self.cached_word_embedding.weight.data[words] = word_reprs.squeeze(1) - - print("Finish generating cached word representations. Going to delete the character encoder.") + + logger.info("Finish generating cached word representations. Going to delete the character encoder.") del self.token_embedder, self.words_to_chars_embedding else: - print("There is no need to cache word representations, since no character information is used.") - + logger.info("There is no need to cache word representations, since no character information is used.") + def forward(self, words): """ @@ -320,7 +325,7 @@ class _ElmoModel(nn.Module): else: chars = None token_embedding = self.token_embedder(expanded_words, chars) # batch_size x max_len x embed_dim - + encoder_output = self.encoder(token_embedding, seq_len) if encoder_output.size(2) < max_len + 2: num_layers, _, output_len, hidden_size = encoder_output.size() @@ -331,7 +336,7 @@ class _ElmoModel(nn.Module): token_embedding = token_embedding.masked_fill(mask, 0) token_embedding = torch.cat((token_embedding, token_embedding), dim=2).view(1, sz[1], sz[2], sz[3]) encoder_output = torch.cat((token_embedding, encoder_output), dim=0) - + # 删除, . 这里没有精确地删除，但应该也不会影响最后的结果了。 encoder_output = encoder_output[:, :, 1:-1] return encoder_output diff --git a/fastNLP/embeddings/embedding.py b/fastNLP/embeddings/embedding.py index 111bacd0..255b0823 100644 --- a/fastNLP/embeddings/embedding.py +++ b/fastNLP/embeddings/embedding.py @@ -3,6 +3,10 @@ """ +__all__ = [ + "Embedding", + "TokenEmbedding" +] import torch.nn as nn from abc import abstractmethod @@ -13,13 +17,12 @@ from .utils import get_embeddings class Embedding(nn.Module): """ - 别名：:class:`fastNLP.embeddings.Embedding` :class:`fastNLP.embeddings.embedding.Embedding` - 词向量嵌入，支持输入多种方式初始化. 可以通过self.num_embeddings获取词表大小; self.embedding_dim获取embedding的维度. Example:: >>> import numpy as np + >>> from fastNLP.embeddings import Embedding >>> init_embed = (2000, 100) >>> embed = Embedding(init_embed) # 随机初始化一个具有2000个词，每个词表示为100维的词向量 >>> init_embed = np.zeros((2000, 100)) @@ -32,54 +35,59 @@ class Embedding(nn.Module): :param float dropout: 对Embedding的输出的dropout。 :param int unk_index: drop word时替换为的index。fastNLP的Vocabulary的unk_index默认为1。 """ - + def __init__(self, init_embed, word_dropout=0, dropout=0.0, unk_index=None): - + super(Embedding, self).__init__() - + self.embed = get_embeddings(init_embed) self.dropout = nn.Dropout(dropout) if not isinstance(self.embed, TokenEmbedding): - self._embed_size = self.embed.weight.size(1) - if word_dropout>0 and not isinstance(unk_index, int): + if hasattr(self.embed, 'embed_size'): + self._embed_size = self.embed.embed_size + elif hasattr(self.embed, 'embedding_dim'): + self._embed_size = self.embed.embedding_dim + else: + self._embed_size = self.embed.weight.size(1) + if word_dropout > 0 and not isinstance(unk_index, int): raise ValueError("When drop word is set, you need to pass in the unk_index.") else: self._embed_size = self.embed.embed_size unk_index = self.embed.get_word_vocab().unknown_idx self.unk_index = unk_index self.word_dropout = word_dropout - + def forward(self, words): """ :param torch.LongTensor words: [batch, seq_len] :return: torch.Tensor : [batch, seq_len, embed_dim] """ - if self.word_dropout>0 and self.training: + if self.word_dropout > 0 and self.training: mask = torch.ones_like(words).float() * self.word_dropout - mask = torch.bernoulli(mask).byte() # dropout_word越大，越多位置为1 + mask = torch.bernoulli(mask).eq(1) # dropout_word越大，越多位置为1 words = words.masked_fill(mask, self.unk_index) words = self.embed(words) return self.dropout(words) - + @property - def num_embedding(self)->int: + def num_embedding(self) -> int: if isinstance(self.embed, nn.Embedding): return self.embed.weight.size(0) else: return self.embed.num_embedding - + def __len__(self): return len(self.embed) - + @property def embed_size(self) -> int: return self._embed_size - + @property def embedding_dim(self) -> int: return self._embed_size - + @property def requires_grad(self): """ @@ -90,14 +98,14 @@ class Embedding(nn.Module): return self.embed.weight.requires_grad else: return self.embed.requires_grad - + @requires_grad.setter def requires_grad(self, value): if not isinstance(self.embed, TokenEmbedding): self.embed.weight.requires_grad = value else: self.embed.requires_grad = value - + @property def size(self): if isinstance(self.embed, TokenEmbedding): @@ -114,12 +122,12 @@ class TokenEmbedding(nn.Module): assert vocab.padding is not None, "Vocabulary must have a padding entry." self._word_vocab = vocab self._word_pad_index = vocab.padding_idx - if word_dropout>0: + if word_dropout > 0: assert vocab.unknown is not None, "Vocabulary must have unknown entry when you want to drop a word." self.word_dropout = word_dropout self._word_unk_index = vocab.unknown_idx self.dropout_layer = nn.Dropout(dropout) - + def drop_word(self, words): """ 按照设定随机将words设置为unknown_index。 @@ -128,11 +136,13 @@ class TokenEmbedding(nn.Module): :return: """ if self.word_dropout > 0 and self.training: - mask = torch.ones_like(words).float() * self.word_dropout - mask = torch.bernoulli(mask).byte() # dropout_word越大，越多位置为1 + mask = torch.full_like(words, fill_value=self.word_dropout, dtype=torch.float, device=words.device) + mask = torch.bernoulli(mask).eq(1) # dropout_word越大，越多位置为1 + pad_mask = words.ne(self._word_pad_index) + mask = mask.__and__(pad_mask) words = words.masked_fill(mask, self._word_unk_index) return words - + def dropout(self, words): """ 对embedding后的word表示进行drop。 @@ -141,7 +151,7 @@ class TokenEmbedding(nn.Module): :return: """ return self.dropout_layer(words) - + @property def requires_grad(self): """ @@ -153,23 +163,23 @@ class TokenEmbedding(nn.Module): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for param in self.parameters(): param.requires_grad = value - + def __len__(self): return len(self._word_vocab) - + @property def embed_size(self) -> int: return self._embed_size - + @property def embedding_dim(self) -> int: return self._embed_size - + @property def num_embedding(self) -> int: """ @@ -177,7 +187,7 @@ class TokenEmbedding(nn.Module): :return: """ return len(self._word_vocab) - + def get_word_vocab(self): """ 返回embedding的词典。 @@ -185,11 +195,11 @@ class TokenEmbedding(nn.Module): :return: Vocabulary """ return self._word_vocab - + @property def size(self): return torch.Size(self.num_embedding, self._embed_size) - + @abstractmethod def forward(self, words): raise NotImplementedError diff --git a/fastNLP/embeddings/stack_embedding.py b/fastNLP/embeddings/stack_embedding.py index 8091d598..e83a275c 100644 --- a/fastNLP/embeddings/stack_embedding.py +++ b/fastNLP/embeddings/stack_embedding.py @@ -1,3 +1,12 @@ +""" +.. todo:: + doc +""" + +__all__ = [ + "StackEmbedding", +] + from typing import List import torch @@ -8,8 +17,6 @@ from .embedding import TokenEmbedding class StackEmbedding(TokenEmbedding): """ - 别名：:class:`fastNLP.embeddings.StackEmbedding` :class:`fastNLP.embeddings.stack_embedding.StackEmbedding` - 支持将多个embedding集合成一个embedding。 Example:: @@ -17,7 +24,7 @@ class StackEmbedding(TokenEmbedding): >>> from fastNLP import Vocabulary >>> from fastNLP.embeddings import StaticEmbedding >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) - >>> embed_1 = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50', requires_grad=True) + >>> embed_1 = StaticEmbedding(vocab, model_dir_or_name='en-glove-6b-50d', requires_grad=True) >>> embed_2 = StaticEmbedding(vocab, model_dir_or_name='en-word2vec-300', requires_grad=True) :param embeds: 一个由若干个TokenEmbedding组成的list，要求每一个TokenEmbedding的词表都保持一致 @@ -26,6 +33,7 @@ class StackEmbedding(TokenEmbedding): :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 """ + def __init__(self, embeds: List[TokenEmbedding], word_dropout=0, dropout=0): vocabs = [] for embed in embeds: @@ -34,14 +42,14 @@ class StackEmbedding(TokenEmbedding): _vocab = vocabs[0] for vocab in vocabs[1:]: assert vocab == _vocab, "All embeddings in StackEmbedding should use the same word vocabulary." - + super(StackEmbedding, self).__init__(_vocab, word_dropout=word_dropout, dropout=dropout) assert isinstance(embeds, list) for embed in embeds: assert isinstance(embed, TokenEmbedding), "Only TokenEmbedding type is supported." self.embeds = nn.ModuleList(embeds) self._embed_size = sum([embed.embed_size for embed in self.embeds]) - + def append(self, embed: TokenEmbedding): """ 添加一个embedding到结尾。 @@ -50,18 +58,18 @@ class StackEmbedding(TokenEmbedding): """ assert isinstance(embed, TokenEmbedding) self.embeds.append(embed) - + def pop(self): """ 弹出最后一个embed :return: """ return self.embeds.pop() - + @property def embed_size(self): return self._embed_size - + @property def requires_grad(self): """ @@ -73,12 +81,12 @@ class StackEmbedding(TokenEmbedding): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for embed in self.embeds(): embed.requires_grad = value - + def forward(self, words): """ 得到多个embedding的结果，并把结果按照顺序concat起来。 @@ -91,4 +99,4 @@ class StackEmbedding(TokenEmbedding): for embed in self.embeds: outputs.append(embed(words)) outputs = self.dropout(torch.cat(outputs, dim=-1)) - return outputs \ No newline at end of file + return outputs diff --git a/fastNLP/embeddings/static_embedding.py b/fastNLP/embeddings/static_embedding.py index 94f7adb5..8249aa11 100644 --- a/fastNLP/embeddings/static_embedding.py +++ b/fastNLP/embeddings/static_embedding.py @@ -1,4 +1,11 @@ +""" +.. todo:: + doc +""" +__all__ = [ + "StaticEmbedding" +] import os import torch @@ -7,25 +14,29 @@ import numpy as np import warnings from ..core.vocabulary import Vocabulary -from ..io.file_utils import PRETRAIN_STATIC_FILES, _get_base_url, cached_path +from ..io.file_utils import PRETRAIN_STATIC_FILES, _get_embedding_url, cached_path from .embedding import TokenEmbedding from ..modules.utils import _get_file_name_base_on_postfix +from copy import deepcopy +from collections import defaultdict +from ..core import logger + class StaticEmbedding(TokenEmbedding): """ - 别名：:class:`fastNLP.embeddings.StaticEmbedding` :class:`fastNLP.embeddings.static_embedding.StaticEmbedding` - StaticEmbedding组件. 给定预训练embedding的名称或路径，根据vocab从embedding中抽取相应的数据(只会将出现在vocab中的词抽取出来， 如果没有找到，则会随机初始化一个值(但如果该word是被标记为no_create_entry的话，则不会单独创建一个值，而是会被指向unk的index))。 当前支持自动下载的预训练vector有以下的几种(待补充); Example:: - + + >>> from fastNLP import Vocabulary + >>> from fastNLP.embeddings import StaticEmbedding >>> vocab = Vocabulary().add_word_lst("The whether is good .".split()) - >>> embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-50') + >>> embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-50d') >>> vocab = Vocabulary().add_word_lst(["The", 'the', "THE"]) - >>> embed = StaticEmbedding(vocab, model_dir_or_name="en-glove-50", lower=True) + >>> embed = StaticEmbedding(vocab, model_dir_or_name="en-glove-50d", lower=True) >>> # "the", "The", "THE"它们共用一个vector，且将使用"the"在预训练词表中寻找它们的初始化表示。 >>> vocab = Vocabulary().add_word_lst(["The", "the", "THE"]) @@ -41,85 +52,120 @@ class StaticEmbedding(TokenEmbedding): :param model_dir_or_name: 可以有两种方式调用预训练好的static embedding：第一种是传入embedding文件夹(文件夹下应该只有一个 以.txt作为后缀的文件)或文件路径；第二种是传入embedding的名称，第二种情况将自动查看缓存中是否存在该模型，没有的话将自动下载。 如果输入为None则使用embedding_dim的维度随机初始化一个embedding。 - :param int embedding_dim: 随机初始化的embedding的维度，仅在model_dir_or_name为None时有效。 + :param int embedding_dim: 随机初始化的embedding的维度，当该值为大于0的值时，将忽略model_dir_or_name。 :param bool requires_grad: 是否需要gradient. 默认为True - :param callable init_method: 如何初始化没有找到的值。可以使用torch.nn.init.*中各种方法。调用该方法时传入一个tensor对象。 + :param callable init_method: 如何初始化没有找到的值。可以使用torch.nn.init.*中各种方法。调用该方法时传入一个tensor对 :param bool lower: 是否将vocab中的词语小写后再和预训练的词表进行匹配。如果你的词表中包含大写的词语，或者就是需要单独 为大写的词语开辟一个vector表示，则将lower设置为False。 - :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 :param float dropout: 以多大的概率对embedding的表示进行Dropout。0.1即随机将10%的值置为0。 + :param float word_dropout: 以多大的概率将一个词替换为unk。这样既可以训练unk也是一定的regularize。 :param bool normalize: 是否对vector进行normalize，使得每个vector的norm为1。 + :param int min_freq: Vocabulary词频数小于这个数量的word将被指向unk。 """ - def __init__(self, vocab: Vocabulary, model_dir_or_name: str='en', embedding_dim=100, requires_grad: bool=True, - init_method=None, lower=False, dropout=0, word_dropout=0, normalize=False): + + def __init__(self, vocab: Vocabulary, model_dir_or_name: str = 'en', embedding_dim=-1, requires_grad: bool = True, + init_method=None, lower=False, dropout=0, word_dropout=0, normalize=False, min_freq=1, **kwargs): super(StaticEmbedding, self).__init__(vocab, word_dropout=word_dropout, dropout=dropout) - + if embedding_dim > 0: + model_dir_or_name = None + # 得到cache_path if model_dir_or_name is None: - assert embedding_dim>=1, "The dimension of embedding should be larger than 1." + assert embedding_dim >= 1, "The dimension of embedding should be larger than 1." embedding_dim = int(embedding_dim) model_path = None elif model_dir_or_name.lower() in PRETRAIN_STATIC_FILES: - PRETRAIN_URL = _get_base_url('static') - model_name = PRETRAIN_STATIC_FILES[model_dir_or_name] - model_url = PRETRAIN_URL + model_name - model_path = cached_path(model_url) + model_url = _get_embedding_url('static', model_dir_or_name.lower()) + model_path = cached_path(model_url, name='embedding') # 检查是否存在 - elif os.path.isfile(os.path.expanduser(os.path.abspath(model_dir_or_name))): - model_path = model_dir_or_name - elif os.path.isdir(os.path.expanduser(os.path.abspath(model_dir_or_name))): - model_path = _get_file_name_base_on_postfix(model_dir_or_name, '.txt') + elif os.path.isfile(os.path.abspath(os.path.expanduser(model_dir_or_name))): + model_path = os.path.abspath(os.path.expanduser(model_dir_or_name)) + elif os.path.isdir(os.path.abspath(os.path.expanduser(model_dir_or_name))): + model_path = _get_file_name_base_on_postfix(os.path.abspath(os.path.expanduser(model_dir_or_name)), '.txt') else: raise ValueError(f"Cannot recognize {model_dir_or_name}.") - + + # 根据min_freq缩小vocab + truncate_vocab = (vocab.min_freq is None and min_freq > 1) or (vocab.min_freq and vocab.min_freq < min_freq) + if truncate_vocab: + truncated_vocab = deepcopy(vocab) + truncated_vocab.min_freq = min_freq + truncated_vocab.word2idx = None + if lower: # 如果有lower，将大小写的的freq需要同时考虑到 + lowered_word_count = defaultdict(int) + for word, count in truncated_vocab.word_count.items(): + lowered_word_count[word.lower()] += count + for word in truncated_vocab.word_count.keys(): + word_count = truncated_vocab.word_count[word] + if lowered_word_count[word.lower()] >= min_freq and word_count < min_freq: + truncated_vocab.add_word_lst([word] * (min_freq - word_count), + no_create_entry=truncated_vocab._is_word_no_create_entry(word)) + + # 只限制在train里面的词语使用min_freq筛选 + if kwargs.get('only_train_min_freq', False) and model_dir_or_name is not None: + for word in truncated_vocab.word_count.keys(): + if truncated_vocab._is_word_no_create_entry(word) and truncated_vocab.word_count[word] < min_freq: + truncated_vocab.add_word_lst([word] * (min_freq - truncated_vocab.word_count[word]), + no_create_entry=True) + truncated_vocab.build_vocab() + truncated_words_to_words = torch.arange(len(vocab)).long() + for word, index in vocab: + truncated_words_to_words[index] = truncated_vocab.to_index(word) + logger.info(f"{len(vocab) - len(truncated_vocab)} out of {len(vocab)} words have frequency less than {min_freq}.") + vocab = truncated_vocab + + self.only_norm_found_vector = kwargs.get('only_norm_found_vector', False) # 读取embedding if lower: lowered_vocab = Vocabulary(padding=vocab.padding, unknown=vocab.unknown) for word, index in vocab: - if not vocab._is_word_no_create_entry(word): + if vocab._is_word_no_create_entry(word): + lowered_vocab.add_word(word.lower(), no_create_entry=True) + else: lowered_vocab.add_word(word.lower()) # 先加入需要创建entry的 - for word in vocab._no_create_word.keys(): # 不需要创建entry的 - if word in vocab: - lowered_word = word.lower() - if lowered_word not in lowered_vocab.word_count: - lowered_vocab.add_word(lowered_word) - lowered_vocab._no_create_word[lowered_word] += 1 - print(f"All word in vocab have been lowered. There are {len(vocab)} words, {len(lowered_vocab)} unique lowered " - f"words.") + logger.info(f"All word in the vocab have been lowered. There are {len(vocab)} words, {len(lowered_vocab)} " + f"unique lowered words.") if model_path: embedding = self._load_with_vocab(model_path, vocab=lowered_vocab, init_method=init_method) else: embedding = self._randomly_init_embed(len(vocab), embedding_dim, init_method) - # 需要适配一下 - if not hasattr(self, 'words_to_words'): - self.words_to_words = torch.arange(len(lowered_vocab, )).long() + self.register_buffer('words_to_words', torch.arange(len(vocab)).long()) if lowered_vocab.unknown: unknown_idx = lowered_vocab.unknown_idx else: unknown_idx = embedding.size(0) - 1 # 否则是最后一个为unknow - words_to_words = nn.Parameter(torch.full((len(vocab),), fill_value=unknown_idx).long(), - requires_grad=False) + self.register_buffer('words_to_words', torch.arange(len(vocab)).long()) + words_to_words = torch.full((len(vocab),), fill_value=unknown_idx).long() for word, index in vocab: if word not in lowered_vocab: word = word.lower() - if lowered_vocab._is_word_no_create_entry(word): # 如果不需要创建entry,已经默认unknown了 - continue + if word not in lowered_vocab and lowered_vocab._is_word_no_create_entry(word): + continue # 如果不需要创建entry,已经默认unknown了 words_to_words[index] = self.words_to_words[lowered_vocab.to_index(word)] - self.words_to_words = words_to_words + self.register_buffer('words_to_words', words_to_words) + self._word_unk_index = lowered_vocab.unknown_idx # 替换一下unknown的index else: if model_path: embedding = self._load_with_vocab(model_path, vocab=vocab, init_method=init_method) else: embedding = self._randomly_init_embed(len(vocab), embedding_dim, init_method) - if normalize: + self.register_buffer('words_to_words', torch.arange(len(vocab)).long()) + if not self.only_norm_found_vector and normalize: embedding /= (torch.norm(embedding, dim=1, keepdim=True) + 1e-12) + + if truncate_vocab: + for i in range(len(truncated_words_to_words)): + index_in_truncated_vocab = truncated_words_to_words[i] + truncated_words_to_words[i] = self.words_to_words[index_in_truncated_vocab] + del self.words_to_words + self.register_buffer('words_to_words', truncated_words_to_words) self.embedding = nn.Embedding(num_embeddings=embedding.shape[0], embedding_dim=embedding.shape[1], padding_idx=vocab.padding_idx, max_norm=None, norm_type=2, scale_grad_by_freq=False, sparse=False, _weight=embedding) self._embed_size = self.embedding.weight.size(1) self.requires_grad = requires_grad - + def _randomly_init_embed(self, num_embedding, embedding_dim, init_embed=None): """ @@ -129,14 +175,14 @@ class StaticEmbedding(TokenEmbedding): :return: torch.FloatTensor """ embed = torch.zeros(num_embedding, embedding_dim) - + if init_embed is None: - nn.init.uniform_(embed, -np.sqrt(3/embedding_dim), np.sqrt(3/embedding_dim)) + nn.init.uniform_(embed, -np.sqrt(3 / embedding_dim), np.sqrt(3 / embedding_dim)) else: init_embed(embed) - + return embed - + @property def requires_grad(self): """ @@ -150,14 +196,14 @@ class StaticEmbedding(TokenEmbedding): return requires_grads.pop() else: return None - + @requires_grad.setter def requires_grad(self, value): for name, param in self.named_parameters(): if 'words_to_words' in name: continue param.requires_grad = value - + def _load_with_vocab(self, embed_filepath, vocab, dtype=np.float32, padding='', unknown='', error='ignore', init_method=None): """ @@ -189,7 +235,12 @@ class StaticEmbedding(TokenEmbedding): dim = len(parts) - 1 f.seek(0) matrix = {} + if vocab.padding: + matrix[vocab.padding_idx] = torch.zeros(dim) + if vocab.unknown: + matrix[vocab.unknown_idx] = torch.zeros(dim) found_count = 0 + found_unknown = False for idx, line in enumerate(f, start_idx): try: parts = line.strip().split() @@ -200,46 +251,42 @@ class StaticEmbedding(TokenEmbedding): word = vocab.padding elif word == unknown and vocab.unknown is not None: word = vocab.unknown + found_unknown = True if word in vocab: index = vocab.to_index(word) matrix[index] = torch.from_numpy(np.fromstring(' '.join(nums), sep=' ', dtype=dtype, count=dim)) + if self.only_norm_found_vector: + matrix[index] = matrix[index] / np.linalg.norm(matrix[index]) found_count += 1 except Exception as e: if error == 'ignore': warnings.warn("Error occurred at the {} line.".format(idx)) else: - print("Error occurred at the {} line.".format(idx)) + logger.error("Error occurred at the {} line.".format(idx)) raise e - print("Found {} out of {} words in the pre-training embedding.".format(found_count, len(vocab))) + logger.info("Found {} out of {} words in the pre-training embedding.".format(found_count, len(vocab))) for word, index in vocab: if index not in matrix and not vocab._is_word_no_create_entry(word): - if vocab.unknown_idx in matrix: # 如果有unkonwn，用unknown初始化 + if found_unknown: # 如果有unkonwn，用unknown初始化 matrix[index] = matrix[vocab.unknown_idx] else: matrix[index] = None - + # matrix中代表是需要建立entry的词 vectors = self._randomly_init_embed(len(matrix), dim, init_method) - - if vocab._no_create_word_length>0: - if vocab.unknown is None: # 创建一个专门的unknown - unknown_idx = len(matrix) - vectors = torch.cat((vectors, torch.zeros(1, dim)), dim=0).contiguous() - else: - unknown_idx = vocab.unknown_idx - words_to_words = nn.Parameter(torch.full((len(vocab),), fill_value=unknown_idx).long(), - requires_grad=False) - for order, (index, vec) in enumerate(matrix.items()): - if vec is not None: - vectors[order] = vec - words_to_words[index] = order - self.words_to_words = words_to_words + + if vocab.unknown is None: # 创建一个专门的unknown + unknown_idx = len(matrix) + vectors = torch.cat((vectors, torch.zeros(1, dim)), dim=0).contiguous() else: - for index, vec in matrix.items(): - if vec is not None: - vectors[index] = vec - + unknown_idx = vocab.unknown_idx + self.register_buffer('words_to_words', torch.full((len(vocab), ), fill_value=unknown_idx).long()) + for index, (index_in_vocab, vec) in enumerate(matrix.items()): + if vec is not None: + vectors[index] = vec + self.words_to_words[index_in_vocab] = index + return vectors - + def forward(self, words): """ 传入words的index diff --git a/fastNLP/embeddings/utils.py b/fastNLP/embeddings/utils.py index b79f563c..844a0c93 100644 --- a/fastNLP/embeddings/utils.py +++ b/fastNLP/embeddings/utils.py @@ -1,13 +1,19 @@ +""" +.. todo:: + doc +""" import numpy as np import torch from torch import nn as nn from ..core.vocabulary import Vocabulary -__all__ = ['get_embeddings'] +__all__ = [ + 'get_embeddings' +] -def _construct_char_vocab_from_vocab(vocab:Vocabulary, min_freq:int=1): +def _construct_char_vocab_from_vocab(vocab: Vocabulary, min_freq: int = 1): """ 给定一个word的vocabulary生成character的vocabulary. @@ -31,13 +37,13 @@ def get_embeddings(init_embed): :param init_embed: 可以是 tuple:(num_embedings, embedding_dim), 即embedding的大小和每个词的维度;也可以传入 nn.Embedding 对象, 此时就以传入的对象作为embedding; 传入np.ndarray也行，将使用传入的ndarray作为作为Embedding初始化; 传入torch.Tensor, 将使用传入的值作为Embedding初始化。 - :return nn.Embedding embeddings: + :return nn.Embedding: embeddings """ if isinstance(init_embed, tuple): res = nn.Embedding( num_embeddings=init_embed[0], embedding_dim=init_embed[1]) - nn.init.uniform_(res.weight.data, a=-np.sqrt(3/res.weight.data.size(1)), - b=np.sqrt(3/res.weight.data.size(1))) + nn.init.uniform_(res.weight.data, a=-np.sqrt(3 / res.weight.data.size(1)), + b=np.sqrt(3 / res.weight.data.size(1))) elif isinstance(init_embed, nn.Module): res = init_embed elif isinstance(init_embed, torch.Tensor): @@ -48,4 +54,4 @@ def get_embeddings(init_embed): else: raise TypeError( 'invalid init_embed type: {}'.format((type(init_embed)))) - return res \ No newline at end of file + return res diff --git a/fastNLP/io/__init__.py b/fastNLP/io/__init__.py index cd0d3527..c8b3dfaa 100644 --- a/fastNLP/io/__init__.py +++ b/fastNLP/io/__init__.py @@ -3,45 +3,92 @@ 1. 用于读入 embedding 的 :doc:`EmbedLoader ` 类, -2. 用于读入不同格式数据的 :doc:`DataSetLoader ` 类 +2. 用于读入不同格式数据的 :doc:`Loader ` 类 -3. 用于读入不同数据集并进行预处理的 :doc:`DataLoader ` 类 +3. 用于处理读入数据的 :doc:`Pipe ` 类 4. 用于保存和载入模型的类, 参考 :doc:`model_io文档` 这些类的使用方法如下: """ __all__ = [ + 'DataBundle', + 'EmbedLoader', + + 'Loader', + + 'YelpLoader', + 'YelpFullLoader', + 'YelpPolarityLoader', + 'IMDBLoader', + 'SSTLoader', + 'SST2Loader', + "ChnSentiCorpLoader", + + 'ConllLoader', + 'Conll2003Loader', + 'Conll2003NERLoader', + 'OntoNotesNERLoader', + 'CTBLoader', + "MsraNERLoader", + "WeiboNERLoader", + "PeopleDailyNERLoader", 'CSVLoader', 'JsonLoader', - 'DataBundle', - 'DataSetLoader', + 'CWSLoader', - 'ConllLoader', - 'Conll2003Loader', - 'IMDBLoader', - 'MatchingLoader', - 'SNLILoader', 'MNLILoader', - 'MTL16Loader', - 'PeopleDailyCorpusLoader', - 'QNLILoader', - 'QuoraLoader', - 'RTELoader', - 'SSTLoader', - 'SST2Loader', - 'YelpLoader', - + "QuoraLoader", + "SNLILoader", + "QNLILoader", + "RTELoader", + + "Pipe", + + "YelpFullPipe", + "YelpPolarityPipe", + "SSTPipe", + "SST2Pipe", + "IMDBPipe", + "ChnSentiCorpPipe", + + "Conll2003Pipe", + "Conll2003NERPipe", + "OntoNotesNERPipe", + "MsraNERPipe", + "PeopleDailyPipe", + "WeiboNERPipe", + + "CWSPipe", + + "MatchingBertPipe", + "RTEBertPipe", + "SNLIBertPipe", + "QuoraBertPipe", + "QNLIBertPipe", + "MNLIBertPipe", + "MatchingPipe", + "RTEPipe", + "SNLIPipe", + "QuoraPipe", + "QNLIPipe", + "MNLIPipe", + 'ModelLoader', 'ModelSaver', + ] from .embed_loader import EmbedLoader -from .base_loader import DataBundle, DataSetLoader -from .dataset_loader import CSVLoader, JsonLoader +from .data_bundle import DataBundle from .model_io import ModelLoader, ModelSaver -from .data_loader import * +from .loader import * +from .pipe import * + +import sys +from ..doc_utils import doc_process +doc_process(sys.modules[__name__]) \ No newline at end of file diff --git a/fastNLP/io/base_loader.py b/fastNLP/io/base_loader.py deleted file mode 100644 index 5d61c16a..00000000 --- a/fastNLP/io/base_loader.py +++ /dev/null @@ -1,220 +0,0 @@ -__all__ = [ - "BaseLoader", - 'DataBundle', - 'DataSetLoader', -] - -import _pickle as pickle -import os -from typing import Union, Dict -import os -from ..core.dataset import DataSet - - -class BaseLoader(object): - """ - 各个 Loader 的基类，提供了 API 的参考。 - - """ - - def __init__(self): - super(BaseLoader, self).__init__() - - @staticmethod - def load_lines(data_path): - """ - 按行读取，舍弃每行两侧空白字符，返回list of str - - :param data_path: 读取数据的路径 - """ - with open(data_path, "r", encoding="utf=8") as f: - text = f.readlines() - return [line.strip() for line in text] - - @classmethod - def load(cls, data_path): - """ - 先按行读取，去除一行两侧空白，再提取每行的字符。返回list of list of str - - :param data_path: - """ - with open(data_path, "r", encoding="utf-8") as f: - text = f.readlines() - return [[word for word in sent.strip()] for sent in text] - - @classmethod - def load_with_cache(cls, data_path, cache_path): - """缓存版的load - """ - if os.path.isfile(cache_path) and os.path.getmtime(data_path) < os.path.getmtime(cache_path): - with open(cache_path, 'rb') as f: - return pickle.load(f) - else: - obj = cls.load(data_path) - with open(cache_path, 'wb') as f: - pickle.dump(obj, f) - return obj - - -def _download_from_url(url, path): - try: - from tqdm.auto import tqdm - except: - from ..core.utils import _pseudo_tqdm as tqdm - import requests - - """Download file""" - r = requests.get(url, headers={'User-Agent': 'Mozilla/5.0'}, stream=True) - chunk_size = 16 * 1024 - total_size = int(r.headers.get('Content-length', 0)) - with open(path, "wb") as file, \ - tqdm(total=total_size, unit='B', unit_scale=1, desc=path.split('/')[-1]) as t: - for chunk in r.iter_content(chunk_size): - if chunk: - file.write(chunk) - t.update(len(chunk)) - - -def _uncompress(src, dst): - import zipfile - import gzip - import tarfile - import os - - def unzip(src, dst): - with zipfile.ZipFile(src, 'r') as f: - f.extractall(dst) - - def ungz(src, dst): - with gzip.open(src, 'rb') as f, open(dst, 'wb') as uf: - length = 16 * 1024 # 16KB - buf = f.read(length) - while buf: - uf.write(buf) - buf = f.read(length) - - def untar(src, dst): - with tarfile.open(src, 'r:gz') as f: - f.extractall(dst) - - fn, ext = os.path.splitext(src) - _, ext_2 = os.path.splitext(fn) - if ext == '.zip': - unzip(src, dst) - elif ext == '.gz' and ext_2 != '.tar': - ungz(src, dst) - elif (ext == '.gz' and ext_2 == '.tar') or ext_2 == '.tgz': - untar(src, dst) - else: - raise ValueError('unsupported file {}'.format(src)) - - -class DataBundle: - """ - 经过处理的数据信息，包括一系列数据集（比如：分开的训练集、验证集和测试集）以及各个field对应的vocabulary。 - - :param vocabs: 从名称(字符串)到 :class:`~fastNLP.Vocabulary` 类型的dict - :param datasets: 从名称(字符串)到 :class:`~fastNLP.DataSet` 类型的dict - """ - - def __init__(self, vocabs: dict = None, datasets: dict = None): - self.vocabs = vocabs or {} - self.datasets = datasets or {} - - def __repr__(self): - _str = 'In total {} datasets:\n'.format(len(self.datasets)) - for name, dataset in self.datasets.items(): - _str += '\t{} has {} instances.\n'.format(name, len(dataset)) - _str += 'In total {} vocabs:\n'.format(len(self.vocabs)) - for name, vocab in self.vocabs.items(): - _str += '\t{} has {} entries.\n'.format(name, len(vocab)) - return _str - - -class DataSetLoader: - """ - 别名：:class:`fastNLP.io.DataSetLoader` :class:`fastNLP.io.dataset_loader.DataSetLoader` - - 定义了各种 DataSetLoader 所需的API 接口，开发者应该继承它实现各种的 DataSetLoader。 - - 开发者至少应该编写如下内容: - - - _load 函数：从一个数据文件中读取数据到一个 :class:`~fastNLP.DataSet` - - load 函数（可以使用基类的方法）：从一个或多个数据文件中读取数据到一个或多个 :class:`~fastNLP.DataSet` - - process 函数：一个或多个从数据文件中读取数据，并处理成可以训练的一个或多个 :class:`~fastNLP.DataSet` - - **process 函数中可以 调用load 函数或 _load 函数** - - """ - URL = '' - DATA_DIR = '' - - ROOT_DIR = '.fastnlp/datasets/' - UNCOMPRESS = True - - def _download(self, url: str, pdir: str, uncompress=True) -> str: - """ - - 从 ``url`` 下载数据到 ``path``， 如果 ``uncompress`` 为 ``True`` ，自动解压。 - - :param url: 下载的网站 - :param pdir: 下载到的目录 - :param uncompress: 是否自动解压缩 - :return: 数据的存放路径 - """ - fn = os.path.basename(url) - path = os.path.join(pdir, fn) - """check data exists""" - if not os.path.exists(path): - os.makedirs(pdir, exist_ok=True) - _download_from_url(url, path) - if uncompress: - dst = os.path.join(pdir, 'data') - if not os.path.exists(dst): - _uncompress(path, dst) - return dst - return path - - def download(self): - return self._download( - self.URL, - os.path.join(self.ROOT_DIR, self.DATA_DIR), - uncompress=self.UNCOMPRESS) - - def load(self, paths: Union[str, Dict[str, str]]) -> Union[DataSet, Dict[str, DataSet]]: - """ - 从指定一个或多个路径中的文件中读取数据，返回一个或多个数据集 :class:`~fastNLP.DataSet` 。 - 如果处理多个路径，传入的 dict 中的 key 与返回的 dict 中的 key 保存一致。 - - :param Union[str, Dict[str, str]] paths: 文件路径 - :return: :class:`~fastNLP.DataSet` 类的对象或存储多个 :class:`~fastNLP.DataSet` 的字典 - """ - if isinstance(paths, str): - return self._load(paths) - return {name: self._load(path) for name, path in paths.items()} - - def _load(self, path: str) -> DataSet: - """从指定路径的文件中读取数据,返回 :class:`~fastNLP.DataSet` 类型的对象 - - :param str path: 文件路径 - :return: 一个 :class:`~fastNLP.DataSet` 类型的对象 - """ - raise NotImplementedError - - def process(self, paths: Union[str, Dict[str, str]], **options) -> DataBundle: - """ - 对于特定的任务和数据集，读取并处理数据，返回处理DataInfo类对象或字典。 - - 从指定一个或多个路径中的文件中读取数据，DataInfo对象中可以包含一个或多个数据集 。 - 如果处理多个路径，传入的 dict 的 key 与返回DataInfo中的 dict 中的 key 保存一致。 - - 返回的 :class:`DataBundle` 对象有如下属性： - - - vocabs: 由从数据集中获取的词表组成的字典，每个词表 - - datasets: 一个dict，包含一系列 :class:`~fastNLP.DataSet` 类型的对象。其中 field 的命名参考 :mod:`~fastNLP.core.const` - - :param paths: 原始数据读取的路径 - :param options: 根据不同的任务和数据集，设计自己的参数 - :return: 返回一个 DataBundle - """ - raise NotImplementedError diff --git a/fastNLP/io/config_io.py b/fastNLP/io/config_io.py deleted file mode 100644 index 4acdbb96..00000000 --- a/fastNLP/io/config_io.py +++ /dev/null @@ -1,311 +0,0 @@ -""" -用于读入和处理和保存 config 文件 - .. todo:: - 这个模块中的类可能被抛弃？ -""" -__all__ = [ - "ConfigLoader", - "ConfigSection", - "ConfigSaver" -] - -import configparser -import json -import os - -from .base_loader import BaseLoader - - -class ConfigLoader(BaseLoader): - """ - 别名：:class:`fastNLP.io.ConfigLoader` :class:`fastNLP.io.config_io.ConfigLoader` - - 读取配置文件的Loader - - :param str data_path: 配置文件的路径 - - """ - - def __init__(self, data_path=None): - super(ConfigLoader, self).__init__() - if data_path is not None: - self.config = self.parse(super(ConfigLoader, self).load(data_path)) - - @staticmethod - def parse(string): - raise NotImplementedError - - @staticmethod - def load_config(file_path, sections): - """ - 把配置文件的section 存入提供的 ``sections`` 中 - - :param str file_path: 配置文件的路径 - :param dict sections: 符合如下键值对组成的字典 `section_name(string)` : :class:`~fastNLP.io.ConfigSection` - - Example:: - - test_args = ConfigSection() - ConfigLoader("config.cfg").load_config("./data_for_tests/config", {"POS_test": test_args}) - - """ - assert isinstance(sections, dict) - cfg = configparser.ConfigParser() - if not os.path.exists(file_path): - raise FileNotFoundError("config file {} not found. ".format(file_path)) - cfg.read(file_path) - for s in sections: - attr_list = [i for i in sections[s].__dict__.keys() if - not callable(getattr(sections[s], i)) and not i.startswith("__")] - if s not in cfg: - print('section %s not found in config file' % (s)) - continue - gen_sec = cfg[s] - for attr in gen_sec.keys(): - try: - val = json.loads(gen_sec[attr]) - # print(s, attr, val, type(val)) - if attr in attr_list: - assert type(val) == type(getattr(sections[s], attr)), \ - 'type not match, except %s but got %s' % \ - (type(getattr(sections[s], attr)), type(val)) - """ - if attr in attr_list then check its type and - update its value. - else add a new attr in sections[s] - """ - setattr(sections[s], attr, val) - except Exception as e: - print("cannot load attribute %s in section %s" - % (attr, s)) - pass - - -class ConfigSection(object): - """ - 别名：:class:`fastNLP.io.ConfigSection` :class:`fastNLP.io.config_io.ConfigSection` - - ConfigSection是一个存储了一个section中所有键值对的数据结构，推荐使用此类的实例来配合 :meth:`ConfigLoader.load_config` 使用 - - """ - - def __init__(self): - super(ConfigSection, self).__init__() - - def __getitem__(self, key): - """ - :param key: str, the name of the attribute - :return attr: the value of this attribute - if key not in self.__dict__.keys(): - return self[key] - else: - raise AttributeError - """ - if key in self.__dict__.keys(): - return getattr(self, key) - raise AttributeError("do NOT have attribute %s" % key) - - def __setitem__(self, key, value): - """ - :param key: str, the name of the attribute - :param value: the value of this attribute - if key not in self.__dict__.keys(): - self[key] will be added - else: - self[key] will be updated - """ - if key in self.__dict__.keys(): - if not isinstance(value, type(getattr(self, key))): - raise AttributeError("attr %s except %s but got %s" % - (key, str(type(getattr(self, key))), str(type(value)))) - setattr(self, key, value) - - def __contains__(self, item): - """ - :param item: The key of item. - :return: True if the key in self.__dict__.keys() else False. - """ - return item in self.__dict__.keys() - - def __eq__(self, other): - """Overwrite the == operator - - :param other: Another ConfigSection() object which to be compared. - :return: True if value of each key in each ConfigSection() object are equal to the other, else False. - """ - for k in self.__dict__.keys(): - if k not in other.__dict__.keys(): - return False - if getattr(self, k) != getattr(self, k): - return False - - for k in other.__dict__.keys(): - if k not in self.__dict__.keys(): - return False - if getattr(self, k) != getattr(self, k): - return False - - return True - - def __ne__(self, other): - """Overwrite the != operator - - :param other: - :return: - """ - return not self.__eq__(other) - - @property - def data(self): - return self.__dict__ - - -class ConfigSaver(object): - """ - 别名：:class:`fastNLP.io.ConfigSaver` :class:`fastNLP.io.config_io.ConfigSaver` - - ConfigSaver 是用来存储配置文件并解决相关冲突的类 - - :param str file_path: 配置文件的路径 - - """ - - def __init__(self, file_path): - self.file_path = file_path - if not os.path.exists(self.file_path): - raise FileNotFoundError("file {} NOT found!".__format__(self.file_path)) - - def _get_section(self, sect_name): - """ - This is the function to get the section with the section name. - - :param sect_name: The name of section what wants to load. - :return: The section. - """ - sect = ConfigSection() - ConfigLoader().load_config(self.file_path, {sect_name: sect}) - return sect - - def _read_section(self): - """ - This is the function to read sections from the config file. - - :return: sect_list, sect_key_list - sect_list: A list of ConfigSection(). - sect_key_list: A list of names in sect_list. - """ - sect_name = None - - sect_list = {} - sect_key_list = [] - - single_section = {} - single_section_key = [] - - with open(self.file_path, 'r') as f: - lines = f.readlines() - - for line in lines: - if line.startswith('[') and line.endswith(']\n'): - if sect_name is None: - pass - else: - sect_list[sect_name] = single_section, single_section_key - single_section = {} - single_section_key = [] - sect_key_list.append(sect_name) - sect_name = line[1: -2] - continue - - if line.startswith('#'): - single_section[line] = '#' - single_section_key.append(line) - continue - - if line.startswith('\n'): - single_section_key.append('\n') - continue - - if '=' not in line: - raise RuntimeError("can NOT load config file {}".__format__(self.file_path)) - - key = line.split('=', maxsplit=1)[0].strip() - value = line.split('=', maxsplit=1)[1].strip() + '\n' - single_section[key] = value - single_section_key.append(key) - - if sect_name is not None: - sect_list[sect_name] = single_section, single_section_key - sect_key_list.append(sect_name) - return sect_list, sect_key_list - - def _write_section(self, sect_list, sect_key_list): - """ - This is the function to write config file with section list and name list. - - :param sect_list: A list of ConfigSection() need to be writen into file. - :param sect_key_list: A list of name of sect_list. - :return: - """ - with open(self.file_path, 'w') as f: - for sect_key in sect_key_list: - single_section, single_section_key = sect_list[sect_key] - f.write('[' + sect_key + ']\n') - for key in single_section_key: - if key == '\n': - f.write('\n') - continue - if single_section[key] == '#': - f.write(key) - continue - f.write(key + ' = ' + single_section[key]) - f.write('\n') - - def save_config_file(self, section_name, section): - """ - 这个方法可以用来修改并保存配置文件中单独的一个 section - - :param str section_name: 需要保存的 section 的名字. - :param section: 你需要修改并保存的 section， :class:`~fastNLP.io.ConfigSaver` 类型 - """ - section_file = self._get_section(section_name) - if len(section_file.__dict__.keys()) == 0: # the section not in the file before - # append this section to config file - with open(self.file_path, 'a') as f: - f.write('[' + section_name + ']\n') - for k in section.__dict__.keys(): - f.write(k + ' = ') - if isinstance(section[k], str): - f.write('\"' + str(section[k]) + '\"\n\n') - else: - f.write(str(section[k]) + '\n\n') - else: - # the section exists - change_file = False - for k in section.__dict__.keys(): - if k not in section_file: - # find a new key in this section - change_file = True - break - if section_file[k] != section[k]: - change_file = True - break - if not change_file: - return - - sect_list, sect_key_list = self._read_section() - if section_name not in sect_key_list: - raise AttributeError() - - sect, sect_key = sect_list[section_name] - for k in section.__dict__.keys(): - if k not in sect_key: - if sect_key[-1] != '\n': - sect_key.append('\n') - sect_key.append(k) - sect[k] = str(section[k]) - if isinstance(section[k], str): - sect[k] = "\"" + sect[k] + "\"" - sect[k] = sect[k] + "\n" - sect_list[section_name] = sect, sect_key - self._write_section(sect_list, sect_key_list) diff --git a/fastNLP/io/data_bundle.py b/fastNLP/io/data_bundle.py new file mode 100644 index 00000000..19b48828 --- /dev/null +++ b/fastNLP/io/data_bundle.py @@ -0,0 +1,320 @@ +""" +.. todo:: + doc +""" +__all__ = [ + 'DataBundle', +] + +from ..core.dataset import DataSet +from ..core.vocabulary import Vocabulary +from typing import Union + +class DataBundle: + """ + 经过处理的数据信息，包括一系列数据集（比如：分开的训练集、验证集和测试集）以及各个field对应的vocabulary。该对象一般由fastNLP中各种 + Loader的load函数生成，可以通过以下的方法获取里面的内容 + + Example:: + + data_bundle = YelpLoader().load({'train':'/path/to/train', 'dev': '/path/to/dev'}) + train_vocabs = data_bundle.vocabs['train'] + train_data = data_bundle.datasets['train'] + dev_data = data_bundle.datasets['train'] + + :param vocabs: 从名称(字符串)到 :class:`~fastNLP.Vocabulary` 类型的dict + :param datasets: 从名称(字符串)到 :class:`~fastNLP.DataSet` 类型的dict + """ + + def __init__(self, vocabs: dict = None, datasets: dict = None): + self.vocabs = vocabs or {} + self.datasets = datasets or {} + + def set_vocab(self, vocab, field_name): + """ + 向DataBunlde中增加vocab + + :param ~fastNLP.Vocabulary vocab: 词表 + :param str field_name: 这个vocab对应的field名称 + :return: self + """ + assert isinstance(vocab, Vocabulary), "Only fastNLP.Vocabulary supports." + self.vocabs[field_name] = vocab + return self + + def set_dataset(self, dataset, name): + """ + + :param ~fastNLP.DataSet dataset: 传递给DataBundle的DataSet + :param str name: dataset的名称 + :return: self + """ + self.datasets[name] = dataset + return self + + def get_dataset(self, name: str) -> DataSet: + """ + 获取名为name的dataset + + :param str name: dataset的名称，一般为'train', 'dev', 'test' + :return: DataSet + """ + return self.datasets[name] + + def delete_dataset(self, name: str): + """ + 删除名为name的DataSet + + :param str name: + :return: self + """ + self.datasets.pop(name, None) + return self + + def get_vocab(self, field_name: str) -> Vocabulary: + """ + 获取field名为field_name对应的vocab + + :param str field_name: 名称 + :return: Vocabulary + """ + return self.vocabs[field_name] + + def delete_vocab(self, field_name: str): + """ + 删除vocab + :param str field_name: + :return: self + """ + self.vocabs.pop(field_name, None) + return self + + def set_input(self, *field_names, flag=True, use_1st_ins_infer_dim_type=True, ignore_miss_dataset=True): + """ + 将field_names中的field设置为input, 对data_bundle中所有的dataset执行该操作:: + + data_bundle.set_input('words', 'seq_len') # 将words和seq_len这两个field的input属性设置为True + data_bundle.set_input('words', flag=False) # 将words这个field的input属性设置为False + + :param str field_names: field的名称 + :param bool flag: 将field_name的input状态设置为flag + :param bool use_1st_ins_infer_dim_type: 如果为True，将不会check该列是否所有数据都是同样的维度，同样的类型。将直接使用第一 + 行的数据进行类型和维度推断本列的数据的类型和维度。 + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :return: self + """ + for field_name in field_names: + for name, dataset in self.datasets.items(): + if not ignore_miss_dataset and not dataset.has_field(field_name): + raise KeyError(f"Field:{field_name} was not found in DataSet:{name}") + if not dataset.has_field(field_name): + continue + else: + dataset.set_input(field_name, flag=flag, use_1st_ins_infer_dim_type=use_1st_ins_infer_dim_type) + return self + + def set_target(self, *field_names, flag=True, use_1st_ins_infer_dim_type=True, ignore_miss_dataset=True): + """ + 将field_names中的field设置为target, 对data_bundle中所有的dataset执行该操作:: + + data_bundle.set_target('target', 'seq_len') # 将words和target这两个field的input属性设置为True + data_bundle.set_target('target', flag=False) # 将target这个field的input属性设置为False + + :param str field_names: field的名称 + :param bool flag: 将field_name的target状态设置为flag + :param bool use_1st_ins_infer_dim_type: 如果为True，将不会check该列是否所有数据都是同样的维度，同样的类型。将直接使用第一 + 行的数据进行类型和维度推断本列的数据的类型和维度。 + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :return: self + """ + for field_name in field_names: + for name, dataset in self.datasets.items(): + if not ignore_miss_dataset and not dataset.has_field(field_name): + raise KeyError(f"Field:{field_name} was not found in DataSet:{name}") + if not dataset.has_field(field_name): + continue + else: + dataset.set_target(field_name, flag=flag, use_1st_ins_infer_dim_type=use_1st_ins_infer_dim_type) + return self + + def set_pad_val(self, field_name, pad_val, ignore_miss_dataset=True): + """ + 将DataBundle中所有的DataSet中名为field_name的Field的padding值设置为pad_val. + + :param str field_name: + :param int pad_val: + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :return: self + """ + for name, dataset in self.datasets.items(): + if dataset.has_field(field_name=field_name): + dataset.set_pad_val(field_name=field_name, pad_val=pad_val) + elif not ignore_miss_dataset: + raise KeyError(f"{field_name} not found DataSet:{name}.") + return self + + def set_ignore_type(self, *field_names, flag=True, ignore_miss_dataset=True): + """ + 将DataBundle中所有的DataSet中名为*field_names的Field的ignore_type设置为flag状态 + + :param str field_names: + :param bool flag: + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :return: self + """ + for name, dataset in self.datasets.items(): + for field_name in field_names: + if dataset.has_field(field_name=field_name): + dataset.set_ignore_type(field_name, flag=flag) + elif not ignore_miss_dataset: + raise KeyError(f"{field_name} not found DataSet:{name}.") + return self + + def copy_field(self, field_name, new_field_name, ignore_miss_dataset=True): + """ + 将DataBundle中所有的DataSet中名为field_name的Field复制一份并命名为叫new_field_name. + + :param str field_name: + :param str new_field_name: + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :return: self + """ + for name, dataset in self.datasets.items(): + if dataset.has_field(field_name=field_name): + dataset.copy_field(field_name=field_name, new_field_name=new_field_name) + elif not ignore_miss_dataset: + raise KeyError(f"{field_name} not found DataSet:{name}.") + return self + + def rename_field(self, field_name, new_field_name, ignore_miss_dataset=True, rename_vocab=True): + """ + 将DataBundle中所有DataSet中名为field_name的field重命名为new_field_name. + + :param str field_name: + :param str new_field_name: + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :param bool rename_vocab: 如果该field同时也存在于vocabs中，会将该field的名称对应修改 + :return: self + """ + for name, dataset in self.datasets.items(): + if dataset.has_field(field_name=field_name): + dataset.rename_field(field_name=field_name, new_field_name=new_field_name) + elif not ignore_miss_dataset: + raise KeyError(f"{field_name} not found DataSet:{name}.") + if rename_vocab: + if field_name in self.vocabs: + self.vocabs[new_field_name] = self.vocabs.pop(field_name) + + return self + + def delete_field(self, field_name, ignore_miss_dataset=True, delete_vocab=True): + """ + 将DataBundle中所有DataSet中名为field_name的field删除掉. + + :param str field_name: + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :param bool delete_vocab: 如果该field也在vocabs中存在，将该值也一并删除 + :return: self + """ + for name, dataset in self.datasets.items(): + if dataset.has_field(field_name=field_name): + dataset.delete_field(field_name=field_name) + elif not ignore_miss_dataset: + raise KeyError(f"{field_name} not found DataSet:{name}.") + if delete_vocab: + if field_name in self.vocabs: + self.vocabs.pop(field_name) + return self + + def iter_datasets(self)->Union[str, DataSet]: + """ + 迭代data_bundle中的DataSet + + Example:: + + for name, dataset in data_bundle.iter_datasets(): + pass + + :return: + """ + for name, dataset in self.datasets.items(): + yield name, dataset + + def iter_vocabs(self)->Union[str, Vocabulary]: + """ + 迭代data_bundle中的DataSet + + Example: + + for field_name, vocab in data_bundle.iter_vocabs(): + pass + + :return: + """ + for field_name, vocab in self.vocabs.items(): + yield field_name, vocab + + def apply_field(self, func, field_name:str, new_field_name:str, ignore_miss_dataset=True, **kwargs): + """ + 对DataBundle中所有的dataset使用apply_field方法 + + :param callable func: input是instance中名为 `field_name` 的field的内容。 + :param str field_name: 传入func的是哪个field。 + :param str new_field_name: 将func返回的内容放入到 `new_field_name` 这个field中，如果名称与已有的field相同，则覆 + 盖之前的field。如果为None则不创建新的field。 + :param bool ignore_miss_dataset: 当某个field名称在某个dataset不存在时，如果为True，则直接忽略该DataSet; + 如果为False，则报错 + :param optional kwargs: 支持输入is_input,is_target,ignore_type + + 1. is_input: bool, 如果为True则将名为 `new_field_name` 的field设置为input + + 2. is_target: bool, 如果为True则将名为 `new_field_name` 的field设置为target + + 3. ignore_type: bool, 如果为True则将名为 `new_field_name` 的field的ignore_type设置为true, 忽略其类型 + """ + for name, dataset in self.datasets.items(): + if dataset.has_field(field_name=field_name): + dataset.apply_field(func=func, field_name=field_name, new_field_name=new_field_name, **kwargs) + elif not ignore_miss_dataset: + raise KeyError(f"{field_name} not found DataSet:{name}.") + return self + + def apply(self, func, new_field_name:str, **kwargs): + """ + 对DataBundle中所有的dataset使用apply方法 + + :param callable func: input是instance中名为 `field_name` 的field的内容。 + :param str new_field_name: 将func返回的内容放入到 `new_field_name` 这个field中，如果名称与已有的field相同，则覆 + 盖之前的field。如果为None则不创建新的field。 + :param optional kwargs: 支持输入is_input,is_target,ignore_type + + 1. is_input: bool, 如果为True则将名为 `new_field_name` 的field设置为input + + 2. is_target: bool, 如果为True则将名为 `new_field_name` 的field设置为target + + 3. ignore_type: bool, 如果为True则将名为 `new_field_name` 的field的ignore_type设置为true, 忽略其类型 + """ + for name, dataset in self.datasets.items(): + dataset.apply(func, new_field_name=new_field_name, **kwargs) + return self + + def __repr__(self): + _str = '' + if len(self.datasets): + _str += 'In total {} datasets:\n'.format(len(self.datasets)) + for name, dataset in self.datasets.items(): + _str += '\t{} has {} instances.\n'.format(name, len(dataset)) + if len(self.vocabs): + _str += 'In total {} vocabs:\n'.format(len(self.vocabs)) + for name, vocab in self.vocabs.items(): + _str += '\t{} has {} entries.\n'.format(name, len(vocab)) + return _str + + diff --git a/fastNLP/io/data_loader/__init__.py b/fastNLP/io/data_loader/__init__.py deleted file mode 100644 index 5d6b08b0..00000000 --- a/fastNLP/io/data_loader/__init__.py +++ /dev/null @@ -1,35 +0,0 @@ -""" -用于读数据集的模块, 可以读取文本分类、序列标注、Matching任务的数据集 - -这些模块的具体介绍如下，您可以通过阅读 :doc:`教程` 来进行了解。 -""" -__all__ = [ - 'ConllLoader', - 'Conll2003Loader', - 'IMDBLoader', - 'MatchingLoader', - 'SNLILoader', - 'MNLILoader', - 'MTL16Loader', - 'PeopleDailyCorpusLoader', - 'QNLILoader', - 'QuoraLoader', - 'RTELoader', - 'SSTLoader', - 'SST2Loader', - 'YelpLoader', -] - - -from .conll import ConllLoader, Conll2003Loader -from .imdb import IMDBLoader -from .matching import MatchingLoader -from .mnli import MNLILoader -from .mtl import MTL16Loader -from .people_daily import PeopleDailyCorpusLoader -from .qnli import QNLILoader -from .quora import QuoraLoader -from .rte import RTELoader -from .snli import SNLILoader -from .sst import SSTLoader, SST2Loader -from .yelp import YelpLoader diff --git a/fastNLP/io/data_loader/conll.py b/fastNLP/io/data_loader/conll.py deleted file mode 100644 index 9b2402a2..00000000 --- a/fastNLP/io/data_loader/conll.py +++ /dev/null @@ -1,73 +0,0 @@ - -from ...core.dataset import DataSet -from ...core.instance import Instance -from ..base_loader import DataSetLoader -from ..file_reader import _read_conll - - -class ConllLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.ConllLoader` :class:`fastNLP.io.data_loader.ConllLoader` - - 读取Conll格式的数据. 数据格式详见 http://conll.cemantix.org/2012/data.html. 数据中以"-DOCSTART-"开头的行将被忽略，因为 - 该符号在conll 2003中被用为文档分割符。 - - 列号从0开始, 每列对应内容为:: - - Column Type - 0 Document ID - 1 Part number - 2 Word number - 3 Word itself - 4 Part-of-Speech - 5 Parse bit - 6 Predicate lemma - 7 Predicate Frameset ID - 8 Word sense - 9 Speaker/Author - 10 Named Entities - 11:N Predicate Arguments - N Coreference - - :param headers: 每一列数据的名称，需为List or Tuple of str。``header`` 与 ``indexes`` 一一对应 - :param indexes: 需要保留的数据列下标，从0开始。若为 ``None`` ，则所有列都保留。Default: ``None`` - :param dropna: 是否忽略非法数据，若 ``False`` ，遇到非法数据时抛出 ``ValueError`` 。Default: ``False`` - """ - - def __init__(self, headers, indexes=None, dropna=False): - super(ConllLoader, self).__init__() - if not isinstance(headers, (list, tuple)): - raise TypeError( - 'invalid headers: {}, should be list of strings'.format(headers)) - self.headers = headers - self.dropna = dropna - if indexes is None: - self.indexes = list(range(len(self.headers))) - else: - if len(indexes) != len(headers): - raise ValueError - self.indexes = indexes - - def _load(self, path): - ds = DataSet() - for idx, data in _read_conll(path, indexes=self.indexes, dropna=self.dropna): - ins = {h: data[i] for i, h in enumerate(self.headers)} - ds.append(Instance(**ins)) - return ds - - -class Conll2003Loader(ConllLoader): - """ - 别名：:class:`fastNLP.io.Conll2003Loader` :class:`fastNLP.io.data_loader.Conll2003Loader` - - 读取Conll2003数据 - - 关于数据集的更多信息,参考: - https://sites.google.com/site/ermasoftware/getting-started/ne-tagging-conll2003-data - """ - - def __init__(self): - headers = [ - 'tokens', 'pos', 'chunks', 'ner', - ] - super(Conll2003Loader, self).__init__(headers=headers) diff --git a/fastNLP/io/data_loader/imdb.py b/fastNLP/io/data_loader/imdb.py deleted file mode 100644 index d3636cde..00000000 --- a/fastNLP/io/data_loader/imdb.py +++ /dev/null @@ -1,99 +0,0 @@ - -from typing import Union, Dict - -from ..embed_loader import EmbeddingOption, EmbedLoader -from ..base_loader import DataSetLoader, DataBundle -from ...core.vocabulary import VocabularyOption, Vocabulary -from ...core.dataset import DataSet -from ...core.instance import Instance -from ...core.const import Const - -from ..utils import get_tokenizer - - -class IMDBLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.IMDBLoader` :class:`fastNLP.io.data_loader.IMDBLoader` - - 读取IMDB数据集，DataSet包含以下fields: - - words: list(str), 需要分类的文本 - - target: str, 文本的标签 - - """ - - def __init__(self): - super(IMDBLoader, self).__init__() - self.tokenizer = get_tokenizer() - - def _load(self, path): - dataset = DataSet() - with open(path, 'r', encoding="utf-8") as f: - for line in f: - line = line.strip() - if not line: - continue - parts = line.split('\t') - target = parts[0] - words = self.tokenizer(parts[1].lower()) - dataset.append(Instance(words=words, target=target)) - - if len(dataset) == 0: - raise RuntimeError(f"{path} has no valid data.") - - return dataset - - def process(self, - paths: Union[str, Dict[str, str]], - src_vocab_opt: VocabularyOption = None, - tgt_vocab_opt: VocabularyOption = None, - char_level_op=False): - - datasets = {} - info = DataBundle() - for name, path in paths.items(): - dataset = self.load(path) - datasets[name] = dataset - - def wordtochar(words): - chars = [] - for word in words: - word = word.lower() - for char in word: - chars.append(char) - chars.append('') - chars.pop() - return chars - - if char_level_op: - for dataset in datasets.values(): - dataset.apply_field(wordtochar, field_name="words", new_field_name='chars') - - datasets["train"], datasets["dev"] = datasets["train"].split(0.1, shuffle=False) - - src_vocab = Vocabulary() if src_vocab_opt is None else Vocabulary(**src_vocab_opt) - src_vocab.from_dataset(datasets['train'], field_name='words') - - src_vocab.index_dataset(*datasets.values(), field_name='words') - - tgt_vocab = Vocabulary(unknown=None, padding=None) \ - if tgt_vocab_opt is None else Vocabulary(**tgt_vocab_opt) - tgt_vocab.from_dataset(datasets['train'], field_name='target') - tgt_vocab.index_dataset(*datasets.values(), field_name='target') - - info.vocabs = { - Const.INPUT: src_vocab, - Const.TARGET: tgt_vocab - } - - info.datasets = datasets - - for name, dataset in info.datasets.items(): - dataset.set_input(Const.INPUT) - dataset.set_target(Const.TARGET) - - return info - - - diff --git a/fastNLP/io/data_loader/matching.py b/fastNLP/io/data_loader/matching.py deleted file mode 100644 index 481b5056..00000000 --- a/fastNLP/io/data_loader/matching.py +++ /dev/null @@ -1,248 +0,0 @@ -import os - -from typing import Union, Dict, List - -from ...core.const import Const -from ...core.vocabulary import Vocabulary -from ..base_loader import DataBundle, DataSetLoader -from ..file_utils import _get_base_url, cached_path, PRETRAINED_BERT_MODEL_DIR -from ...modules.encoder.bert import BertTokenizer - - -class MatchingLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.MatchingLoader` :class:`fastNLP.io.data_loader.MatchingLoader` - - 读取Matching任务的数据集 - - :param dict paths: key是数据集名称（如train、dev、test），value是对应的文件名 - """ - - def __init__(self, paths: dict=None): - self.paths = paths - - def _load(self, path): - """ - :param str path: 待读取数据集的路径名 - :return: fastNLP.DataSet ds: 返回一个DataSet对象，里面必须包含3个field：其中两个分别为两个句子 - 的原始字符串文本，第三个为标签 - """ - raise NotImplementedError - - def process(self, paths: Union[str, Dict[str, str]], dataset_name: str=None, - to_lower=False, seq_len_type: str=None, bert_tokenizer: str=None, - cut_text: int = None, get_index=True, auto_pad_length: int=None, - auto_pad_token: str='', set_input: Union[list, str, bool]=True, - set_target: Union[list, str, bool]=True, concat: Union[str, list, bool]=None, - extra_split: List[str]=None, ) -> DataBundle: - """ - :param paths: str或者Dict[str, str]。如果是str，则为数据集所在的文件夹或者是全路径文件名：如果是文件夹， - 则会从self.paths里面找对应的数据集名称与文件名。如果是Dict，则为数据集名称（如train、dev、test）和 - 对应的全路径文件名。 - :param str dataset_name: 如果在paths里传入的是一个数据集的全路径文件名，那么可以用dataset_name来定义 - 这个数据集的名字，如果不定义则默认为train。 - :param bool to_lower: 是否将文本自动转为小写。默认值为False。 - :param str seq_len_type: 提供的seq_len类型，支持 ``seq_len`` ：提供一个数字作为句子长度； ``mask`` : - 提供一个0/1的mask矩阵作为句子长度； ``bert`` ：提供segment_type_id（第一个句子为0，第二个句子为1）和 - attention mask矩阵（0/1的mask矩阵）。默认值为None，即不提供seq_len - :param str bert_tokenizer: bert tokenizer所使用的词表所在的文件夹路径 - :param int cut_text: 将长于cut_text的内容截掉。默认为None，即不截。 - :param bool get_index: 是否需要根据词表将文本转为index - :param int auto_pad_length: 是否需要将文本自动pad到一定长度（超过这个长度的文本将会被截掉），默认为不会自动pad - :param str auto_pad_token: 自动pad的内容 - :param set_input: 如果为True，则会自动将相关的field（名字里含有Const.INPUT的）设置为input，如果为False - 则不会将任何field设置为input。如果传入str或者List[str]，则会根据传入的内容将相对应的field设置为input， - 于此同时其他field不会被设置为input。默认值为True。 - :param set_target: set_target将控制哪些field可以被设置为target，用法与set_input一致。默认值为True。 - :param concat: 是否需要将两个句子拼接起来。如果为False则不会拼接。如果为True则会在两个句子之间插入一个。 - 如果传入一个长度为4的list，则分别表示插在第一句开始前、第一句结束后、第二句开始前、第二句结束后的标识符。如果 - 传入字符串 ``bert`` ，则会采用bert的拼接方式，等价于['[CLS]', '[SEP]', '', '[SEP]']. - :param extra_split: 额外的分隔符，即除了空格之外的用于分词的字符。 - :return: - """ - if isinstance(set_input, str): - set_input = [set_input] - if isinstance(set_target, str): - set_target = [set_target] - if isinstance(set_input, bool): - auto_set_input = set_input - else: - auto_set_input = False - if isinstance(set_target, bool): - auto_set_target = set_target - else: - auto_set_target = False - if isinstance(paths, str): - if os.path.isdir(paths): - path = {n: os.path.join(paths, self.paths[n]) for n in self.paths.keys()} - else: - path = {dataset_name if dataset_name is not None else 'train': paths} - else: - path = paths - - data_info = DataBundle() - for data_name in path.keys(): - data_info.datasets[data_name] = self._load(path[data_name]) - - for data_name, data_set in data_info.datasets.items(): - if auto_set_input: - data_set.set_input(Const.INPUTS(0), Const.INPUTS(1)) - if auto_set_target: - if Const.TARGET in data_set.get_field_names(): - data_set.set_target(Const.TARGET) - - if extra_split is not None: - for data_name, data_set in data_info.datasets.items(): - data_set.apply(lambda x: ' '.join(x[Const.INPUTS(0)]), new_field_name=Const.INPUTS(0)) - data_set.apply(lambda x: ' '.join(x[Const.INPUTS(1)]), new_field_name=Const.INPUTS(1)) - - for s in extra_split: - data_set.apply(lambda x: x[Const.INPUTS(0)].replace(s, ' ' + s + ' '), - new_field_name=Const.INPUTS(0)) - data_set.apply(lambda x: x[Const.INPUTS(0)].replace(s, ' ' + s + ' '), - new_field_name=Const.INPUTS(0)) - - _filt = lambda x: x - data_set.apply(lambda x: list(filter(_filt, x[Const.INPUTS(0)].split(' '))), - new_field_name=Const.INPUTS(0), is_input=auto_set_input) - data_set.apply(lambda x: list(filter(_filt, x[Const.INPUTS(1)].split(' '))), - new_field_name=Const.INPUTS(1), is_input=auto_set_input) - _filt = None - - if to_lower: - for data_name, data_set in data_info.datasets.items(): - data_set.apply(lambda x: [w.lower() for w in x[Const.INPUTS(0)]], new_field_name=Const.INPUTS(0), - is_input=auto_set_input) - data_set.apply(lambda x: [w.lower() for w in x[Const.INPUTS(1)]], new_field_name=Const.INPUTS(1), - is_input=auto_set_input) - - if bert_tokenizer is not None: - if bert_tokenizer.lower() in PRETRAINED_BERT_MODEL_DIR: - PRETRAIN_URL = _get_base_url('bert') - model_name = PRETRAINED_BERT_MODEL_DIR[bert_tokenizer] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(bert_tokenizer): - model_dir = bert_tokenizer - else: - raise ValueError(f"Cannot recognize BERT tokenizer from {bert_tokenizer}.") - - words_vocab = Vocabulary(padding='[PAD]', unknown='[UNK]') - with open(os.path.join(model_dir, 'vocab.txt'), 'r') as f: - lines = f.readlines() - lines = [line.strip() for line in lines] - words_vocab.add_word_lst(lines) - words_vocab.build_vocab() - - tokenizer = BertTokenizer.from_pretrained(model_dir) - - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: tokenizer.tokenize(' '.join(x[fields])), new_field_name=fields, - is_input=auto_set_input) - - if isinstance(concat, bool): - concat = 'default' if concat else None - if concat is not None: - if isinstance(concat, str): - CONCAT_MAP = {'bert': ['[CLS]', '[SEP]', '', '[SEP]'], - 'default': ['', '', '', '']} - if concat.lower() in CONCAT_MAP: - concat = CONCAT_MAP[concat] - else: - concat = 4 * [concat] - assert len(concat) == 4, \ - f'Please choose a list with 4 symbols which at the beginning of first sentence ' \ - f'the end of first sentence, the begin of second sentence, and the end of second' \ - f'sentence. Your input is {concat}' - - for data_name, data_set in data_info.datasets.items(): - data_set.apply(lambda x: [concat[0]] + x[Const.INPUTS(0)] + [concat[1]] + [concat[2]] + - x[Const.INPUTS(1)] + [concat[3]], new_field_name=Const.INPUT) - data_set.apply(lambda x: [w for w in x[Const.INPUT] if len(w) > 0], new_field_name=Const.INPUT, - is_input=auto_set_input) - - if seq_len_type is not None: - if seq_len_type == 'seq_len': # - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: len(x[fields]), - new_field_name=fields.replace(Const.INPUT, Const.INPUT_LEN), - is_input=auto_set_input) - elif seq_len_type == 'mask': - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: [1] * len(x[fields]), - new_field_name=fields.replace(Const.INPUT, Const.INPUT_LEN), - is_input=auto_set_input) - elif seq_len_type == 'bert': - for data_name, data_set in data_info.datasets.items(): - if Const.INPUT not in data_set.get_field_names(): - raise KeyError(f'Field ``{Const.INPUT}`` not in {data_name} data set: ' - f'got {data_set.get_field_names()}') - data_set.apply(lambda x: [0] * (len(x[Const.INPUTS(0)]) + 2) + [1] * (len(x[Const.INPUTS(1)]) + 1), - new_field_name=Const.INPUT_LENS(0), is_input=auto_set_input) - data_set.apply(lambda x: [1] * len(x[Const.INPUT_LENS(0)]), - new_field_name=Const.INPUT_LENS(1), is_input=auto_set_input) - - if auto_pad_length is not None: - cut_text = min(auto_pad_length, cut_text if cut_text is not None else auto_pad_length) - - if cut_text is not None: - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if (Const.INPUT in fields) or ((Const.INPUT_LEN in fields) and (seq_len_type != 'seq_len')): - data_set.apply(lambda x: x[fields][: cut_text], new_field_name=fields, - is_input=auto_set_input) - - data_set_list = [d for n, d in data_info.datasets.items()] - assert len(data_set_list) > 0, f'There are NO data sets in data info!' - - if bert_tokenizer is None: - words_vocab = Vocabulary(padding=auto_pad_token) - words_vocab = words_vocab.from_dataset(*[d for n, d in data_info.datasets.items() if 'train' in n], - field_name=[n for n in data_set_list[0].get_field_names() - if (Const.INPUT in n)], - no_create_entry_dataset=[d for n, d in data_info.datasets.items() - if 'train' not in n]) - target_vocab = Vocabulary(padding=None, unknown=None) - target_vocab = target_vocab.from_dataset(*[d for n, d in data_info.datasets.items() if 'train' in n], - field_name=Const.TARGET) - data_info.vocabs = {Const.INPUT: words_vocab, Const.TARGET: target_vocab} - - if get_index: - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: [words_vocab.to_index(w) for w in x[fields]], new_field_name=fields, - is_input=auto_set_input) - - if Const.TARGET in data_set.get_field_names(): - data_set.apply(lambda x: target_vocab.to_index(x[Const.TARGET]), new_field_name=Const.TARGET, - is_input=auto_set_input, is_target=auto_set_target) - - if auto_pad_length is not None: - if seq_len_type == 'seq_len': - raise RuntimeError(f'the sequence will be padded with the length {auto_pad_length}, ' - f'so the seq_len_type cannot be `{seq_len_type}`!') - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: x[fields] + [words_vocab.to_index(words_vocab.padding)] * - (auto_pad_length - len(x[fields])), new_field_name=fields, - is_input=auto_set_input) - elif (Const.INPUT_LEN in fields) and (seq_len_type != 'seq_len'): - data_set.apply(lambda x: x[fields] + [0] * (auto_pad_length - len(x[fields])), - new_field_name=fields, is_input=auto_set_input) - - for data_name, data_set in data_info.datasets.items(): - if isinstance(set_input, list): - data_set.set_input(*[inputs for inputs in set_input if inputs in data_set.get_field_names()]) - if isinstance(set_target, list): - data_set.set_target(*[target for target in set_target if target in data_set.get_field_names()]) - - return data_info diff --git a/fastNLP/io/data_loader/mnli.py b/fastNLP/io/data_loader/mnli.py deleted file mode 100644 index 65863f3d..00000000 --- a/fastNLP/io/data_loader/mnli.py +++ /dev/null @@ -1,62 +0,0 @@ - -from ...core.const import Const - -from .matching import MatchingLoader -from ..dataset_loader import CSVLoader - - -class MNLILoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.MNLILoader` :class:`fastNLP.io.data_loader.MNLILoader` - - 读取MNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - - words2: list(str), 第二句文本, hypothesis - - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev_matched': 'dev_matched.tsv', - 'dev_mismatched': 'dev_mismatched.tsv', - 'test_matched': 'test_matched.tsv', - 'test_mismatched': 'test_mismatched.tsv', - # 'test_0.9_matched': 'multinli_0.9_test_matched_unlabeled.txt', - # 'test_0.9_mismatched': 'multinli_0.9_test_mismatched_unlabeled.txt', - - # test_0.9_mathed与mismatched是MNLI0.9版本的（数据来源：kaggle） - } - MatchingLoader.__init__(self, paths=paths) - CSVLoader.__init__(self, sep='\t') - self.fields = { - 'sentence1_binary_parse': Const.INPUTS(0), - 'sentence2_binary_parse': Const.INPUTS(1), - 'gold_label': Const.TARGET, - } - - def _load(self, path): - ds = CSVLoader._load(self, path) - - for k, v in self.fields.items(): - if k in ds.get_field_names(): - ds.rename_field(k, v) - - if Const.TARGET in ds.get_field_names(): - if ds[0][Const.TARGET] == 'hidden': - ds.delete_field(Const.TARGET) - - parentheses_table = str.maketrans({'(': None, ')': None}) - - ds.apply(lambda ins: ins[Const.INPUTS(0)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(0)) - ds.apply(lambda ins: ins[Const.INPUTS(1)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(1)) - if Const.TARGET in ds.get_field_names(): - ds.drop(lambda x: x[Const.TARGET] == '-') - return ds diff --git a/fastNLP/io/data_loader/mtl.py b/fastNLP/io/data_loader/mtl.py deleted file mode 100644 index cbca413d..00000000 --- a/fastNLP/io/data_loader/mtl.py +++ /dev/null @@ -1,68 +0,0 @@ - -from typing import Union, Dict - -from ..base_loader import DataBundle -from ..dataset_loader import CSVLoader -from ...core.vocabulary import Vocabulary, VocabularyOption -from ...core.const import Const -from ..utils import check_dataloader_paths - - -class MTL16Loader(CSVLoader): - """ - 别名：:class:`fastNLP.io.MTL16Loader` :class:`fastNLP.io.data_loader.MTL16Loader` - - 读取MTL16数据集，DataSet包含以下fields: - - words: list(str), 需要分类的文本 - - target: str, 文本的标签 - - 数据来源：https://pan.baidu.com/s/1c2L6vdA - - """ - - def __init__(self): - super(MTL16Loader, self).__init__(headers=(Const.TARGET, Const.INPUT), sep='\t') - - def _load(self, path): - dataset = super(MTL16Loader, self)._load(path) - dataset.apply(lambda x: x[Const.INPUT].lower().split(), new_field_name=Const.INPUT) - if len(dataset) == 0: - raise RuntimeError(f"{path} has no valid data.") - - return dataset - - def process(self, - paths: Union[str, Dict[str, str]], - src_vocab_opt: VocabularyOption = None, - tgt_vocab_opt: VocabularyOption = None,): - - paths = check_dataloader_paths(paths) - datasets = {} - info = DataBundle() - for name, path in paths.items(): - dataset = self.load(path) - datasets[name] = dataset - - src_vocab = Vocabulary() if src_vocab_opt is None else Vocabulary(**src_vocab_opt) - src_vocab.from_dataset(datasets['train'], field_name=Const.INPUT) - src_vocab.index_dataset(*datasets.values(), field_name=Const.INPUT) - - tgt_vocab = Vocabulary(unknown=None, padding=None) \ - if tgt_vocab_opt is None else Vocabulary(**tgt_vocab_opt) - tgt_vocab.from_dataset(datasets['train'], field_name=Const.TARGET) - tgt_vocab.index_dataset(*datasets.values(), field_name=Const.TARGET) - - info.vocabs = { - Const.INPUT: src_vocab, - Const.TARGET: tgt_vocab - } - - info.datasets = datasets - - for name, dataset in info.datasets.items(): - dataset.set_input(Const.INPUT) - dataset.set_target(Const.TARGET) - - return info diff --git a/fastNLP/io/data_loader/people_daily.py b/fastNLP/io/data_loader/people_daily.py deleted file mode 100644 index 5efadb7d..00000000 --- a/fastNLP/io/data_loader/people_daily.py +++ /dev/null @@ -1,85 +0,0 @@ - -from ..base_loader import DataSetLoader -from ...core.dataset import DataSet -from ...core.instance import Instance -from ...core.const import Const - - -class PeopleDailyCorpusLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.PeopleDailyCorpusLoader` :class:`fastNLP.io.data_loader.PeopleDailyCorpusLoader` - - 读取人民日报数据集 - """ - - def __init__(self, pos=True, ner=True): - super(PeopleDailyCorpusLoader, self).__init__() - self.pos = pos - self.ner = ner - - def _load(self, data_path): - with open(data_path, "r", encoding="utf-8") as f: - sents = f.readlines() - examples = [] - for sent in sents: - if len(sent) <= 2: - continue - inside_ne = False - sent_pos_tag = [] - sent_words = [] - sent_ner = [] - words = sent.strip().split()[1:] - for word in words: - if "[" in word and "]" in word: - ner_tag = "U" - print(word) - elif "[" in word: - inside_ne = True - ner_tag = "B" - word = word[1:] - elif "]" in word: - ner_tag = "L" - word = word[:word.index("]")] - if inside_ne is True: - inside_ne = False - else: - raise RuntimeError("only ] appears!") - else: - if inside_ne is True: - ner_tag = "I" - else: - ner_tag = "O" - tmp = word.split("/") - token, pos = tmp[0], tmp[1] - sent_ner.append(ner_tag) - sent_pos_tag.append(pos) - sent_words.append(token) - example = [sent_words] - if self.pos is True: - example.append(sent_pos_tag) - if self.ner is True: - example.append(sent_ner) - examples.append(example) - return self.convert(examples) - - def convert(self, data): - """ - - :param data: python 内置对象 - :return: 一个 :class:`~fastNLP.DataSet` 类型的对象 - """ - data_set = DataSet() - for item in data: - sent_words = item[0] - if self.pos is True and self.ner is True: - instance = Instance( - words=sent_words, pos_tags=item[1], ner=item[2]) - elif self.pos is True: - instance = Instance(words=sent_words, pos_tags=item[1]) - elif self.ner is True: - instance = Instance(words=sent_words, ner=item[1]) - else: - instance = Instance(words=sent_words) - data_set.append(instance) - data_set.apply(lambda ins: len(ins[Const.INPUT]), new_field_name=Const.INPUT_LEN) - return data_set diff --git a/fastNLP/io/data_loader/qnli.py b/fastNLP/io/data_loader/qnli.py deleted file mode 100644 index 84b0f3d6..00000000 --- a/fastNLP/io/data_loader/qnli.py +++ /dev/null @@ -1,47 +0,0 @@ - -from ...core.const import Const - -from .matching import MatchingLoader -from ..dataset_loader import CSVLoader - - -class QNLILoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.QNLILoader` :class:`fastNLP.io.data_loader.QNLILoader` - - 读取QNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - - words2: list(str), 第二句文本, hypothesis - - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev': 'dev.tsv', - 'test': 'test.tsv' # test set has not label - } - MatchingLoader.__init__(self, paths=paths) - self.fields = { - 'question': Const.INPUTS(0), - 'sentence': Const.INPUTS(1), - 'label': Const.TARGET, - } - CSVLoader.__init__(self, sep='\t') - - def _load(self, path): - ds = CSVLoader._load(self, path) - - for k, v in self.fields.items(): - if k in ds.get_field_names(): - ds.rename_field(k, v) - for fields in ds.get_all_fields(): - if Const.INPUT in fields: - ds.apply(lambda x: x[fields].strip().split(), new_field_name=fields) - - return ds diff --git a/fastNLP/io/data_loader/quora.py b/fastNLP/io/data_loader/quora.py deleted file mode 100644 index d0ee41ec..00000000 --- a/fastNLP/io/data_loader/quora.py +++ /dev/null @@ -1,34 +0,0 @@ - -from ...core.const import Const - -from .matching import MatchingLoader -from ..dataset_loader import CSVLoader - - -class QuoraLoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.QuoraLoader` :class:`fastNLP.io.data_loader.QuoraLoader` - - 读取MNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - - words2: list(str), 第二句文本, hypothesis - - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev': 'dev.tsv', - 'test': 'test.tsv', - } - MatchingLoader.__init__(self, paths=paths) - CSVLoader.__init__(self, sep='\t', headers=(Const.TARGET, Const.INPUTS(0), Const.INPUTS(1), 'pairID')) - - def _load(self, path): - ds = CSVLoader._load(self, path) - return ds diff --git a/fastNLP/io/data_loader/rte.py b/fastNLP/io/data_loader/rte.py deleted file mode 100644 index f8c5e2fc..00000000 --- a/fastNLP/io/data_loader/rte.py +++ /dev/null @@ -1,47 +0,0 @@ - -from ...core.const import Const - -from .matching import MatchingLoader -from ..dataset_loader import CSVLoader - - -class RTELoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.RTELoader` :class:`fastNLP.io.data_loader.RTELoader` - - 读取RTE数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - - words2: list(str), 第二句文本, hypothesis - - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev': 'dev.tsv', - 'test': 'test.tsv' # test set has not label - } - MatchingLoader.__init__(self, paths=paths) - self.fields = { - 'sentence1': Const.INPUTS(0), - 'sentence2': Const.INPUTS(1), - 'label': Const.TARGET, - } - CSVLoader.__init__(self, sep='\t') - - def _load(self, path): - ds = CSVLoader._load(self, path) - - for k, v in self.fields.items(): - if k in ds.get_field_names(): - ds.rename_field(k, v) - for fields in ds.get_all_fields(): - if Const.INPUT in fields: - ds.apply(lambda x: x[fields].strip().split(), new_field_name=fields) - - return ds diff --git a/fastNLP/io/data_loader/snli.py b/fastNLP/io/data_loader/snli.py deleted file mode 100644 index 1db0ac5b..00000000 --- a/fastNLP/io/data_loader/snli.py +++ /dev/null @@ -1,46 +0,0 @@ - -from ...core.const import Const - -from .matching import MatchingLoader -from ..dataset_loader import JsonLoader - - -class SNLILoader(MatchingLoader, JsonLoader): - """ - 别名：:class:`fastNLP.io.SNLILoader` :class:`fastNLP.io.data_loader.SNLILoader` - - 读取SNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - - words2: list(str), 第二句文本, hypothesis - - target: str, 真实标签 - - 数据来源: https://nlp.stanford.edu/projects/snli/snli_1.0.zip - """ - - def __init__(self, paths: dict=None): - fields = { - 'sentence1_binary_parse': Const.INPUTS(0), - 'sentence2_binary_parse': Const.INPUTS(1), - 'gold_label': Const.TARGET, - } - paths = paths if paths is not None else { - 'train': 'snli_1.0_train.jsonl', - 'dev': 'snli_1.0_dev.jsonl', - 'test': 'snli_1.0_test.jsonl'} - MatchingLoader.__init__(self, paths=paths) - JsonLoader.__init__(self, fields=fields) - - def _load(self, path): - ds = JsonLoader._load(self, path) - - parentheses_table = str.maketrans({'(': None, ')': None}) - - ds.apply(lambda ins: ins[Const.INPUTS(0)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(0)) - ds.apply(lambda ins: ins[Const.INPUTS(1)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(1)) - ds.drop(lambda x: x[Const.TARGET] == '-') - return ds diff --git a/fastNLP/io/data_loader/sst.py b/fastNLP/io/data_loader/sst.py deleted file mode 100644 index 0d881e65..00000000 --- a/fastNLP/io/data_loader/sst.py +++ /dev/null @@ -1,177 +0,0 @@ - -from typing import Union, Dict -from nltk import Tree - -from ..base_loader import DataBundle, DataSetLoader -from ..dataset_loader import CSVLoader -from ...core.vocabulary import VocabularyOption, Vocabulary -from ...core.dataset import DataSet -from ...core.const import Const -from ...core.instance import Instance -from ..utils import check_dataloader_paths, get_tokenizer - - -class SSTLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.SSTLoader` :class:`fastNLP.io.data_loader.SSTLoader` - - 读取SST数据集, DataSet包含fields:: - - words: list(str) 需要分类的文本 - target: str 文本的标签 - - 数据来源: https://nlp.stanford.edu/sentiment/trainDevTestTrees_PTB.zip - - :param subtree: 是否将数据展开为子树，扩充数据量. Default: ``False`` - :param fine_grained: 是否使用SST-5标准，若 ``False`` , 使用SST-2。Default: ``False`` - """ - - URL = 'https://nlp.stanford.edu/sentiment/trainDevTestTrees_PTB.zip' - DATA_DIR = 'sst/' - - def __init__(self, subtree=False, fine_grained=False): - self.subtree = subtree - - tag_v = {'0': 'very negative', '1': 'negative', '2': 'neutral', - '3': 'positive', '4': 'very positive'} - if not fine_grained: - tag_v['0'] = tag_v['1'] - tag_v['4'] = tag_v['3'] - self.tag_v = tag_v - self.tokenizer = get_tokenizer() - - def _load(self, path): - """ - - :param str path: 存储数据的路径 - :return: 一个 :class:`~fastNLP.DataSet` 类型的对象 - """ - datalist = [] - with open(path, 'r', encoding='utf-8') as f: - datas = [] - for l in f: - datas.extend([(s, self.tag_v[t]) - for s, t in self._get_one(l, self.subtree)]) - ds = DataSet() - for words, tag in datas: - ds.append(Instance(words=words, target=tag)) - return ds - - def _get_one(self, data, subtree): - tree = Tree.fromstring(data) - if subtree: - return [(self.tokenizer(' '.join(t.leaves())), t.label()) for t in tree.subtrees() ] - return [(self.tokenizer(' '.join(tree.leaves())), tree.label())] - - def process(self, - paths, train_subtree=True, - src_vocab_op: VocabularyOption = None, - tgt_vocab_op: VocabularyOption = None,): - paths = check_dataloader_paths(paths) - input_name, target_name = 'words', 'target' - src_vocab = Vocabulary() if src_vocab_op is None else Vocabulary(**src_vocab_op) - tgt_vocab = Vocabulary(unknown=None, padding=None) \ - if tgt_vocab_op is None else Vocabulary(**tgt_vocab_op) - - info = DataBundle() - origin_subtree = self.subtree - self.subtree = train_subtree - info.datasets['train'] = self._load(paths['train']) - self.subtree = origin_subtree - for n, p in paths.items(): - if n != 'train': - info.datasets[n] = self._load(p) - - src_vocab.from_dataset( - info.datasets['train'], - field_name=input_name, - no_create_entry_dataset=[ds for n, ds in info.datasets.items() if n != 'train']) - tgt_vocab.from_dataset(info.datasets['train'], field_name=target_name) - - src_vocab.index_dataset( - *info.datasets.values(), - field_name=input_name, new_field_name=input_name) - tgt_vocab.index_dataset( - *info.datasets.values(), - field_name=target_name, new_field_name=target_name) - info.vocabs = { - input_name: src_vocab, - target_name: tgt_vocab - } - - return info - - -class SST2Loader(CSVLoader): - """ - 别名：:class:`fastNLP.io.SST2Loader` :class:`fastNLP.io.data_loader.SST2Loader` - - 数据来源 SST: https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2FSST-2.zip?alt=media&token=aabc5f6b-e466-44a2-b9b4-cf6337f84ac8 - """ - - def __init__(self): - super(SST2Loader, self).__init__(sep='\t') - self.tokenizer = get_tokenizer() - self.field = {'sentence': Const.INPUT, 'label': Const.TARGET} - - def _load(self, path: str) -> DataSet: - ds = super(SST2Loader, self)._load(path) - for k, v in self.field.items(): - if k in ds.get_field_names(): - ds.rename_field(k, v) - ds.apply(lambda x: self.tokenizer(x[Const.INPUT]), new_field_name=Const.INPUT) - print("all count:", len(ds)) - return ds - - def process(self, - paths: Union[str, Dict[str, str]], - src_vocab_opt: VocabularyOption = None, - tgt_vocab_opt: VocabularyOption = None, - char_level_op=False): - - paths = check_dataloader_paths(paths) - datasets = {} - info = DataBundle() - for name, path in paths.items(): - dataset = self.load(path) - datasets[name] = dataset - - def wordtochar(words): - chars = [] - for word in words: - word = word.lower() - for char in word: - chars.append(char) - chars.append('') - chars.pop() - return chars - - input_name, target_name = Const.INPUT, Const.TARGET - info.vocabs={} - - # 就分隔为char形式 - if char_level_op: - for dataset in datasets.values(): - dataset.apply_field(wordtochar, field_name=Const.INPUT, new_field_name=Const.CHAR_INPUT) - src_vocab = Vocabulary() if src_vocab_opt is None else Vocabulary(**src_vocab_opt) - src_vocab.from_dataset(datasets['train'], field_name=Const.INPUT) - src_vocab.index_dataset(*datasets.values(), field_name=Const.INPUT) - - tgt_vocab = Vocabulary(unknown=None, padding=None) \ - if tgt_vocab_opt is None else Vocabulary(**tgt_vocab_opt) - tgt_vocab.from_dataset(datasets['train'], field_name=Const.TARGET) - tgt_vocab.index_dataset(*datasets.values(), field_name=Const.TARGET) - - info.vocabs = { - Const.INPUT: src_vocab, - Const.TARGET: tgt_vocab - } - - info.datasets = datasets - - for name, dataset in info.datasets.items(): - dataset.set_input(Const.INPUT) - dataset.set_target(Const.TARGET) - - return info - diff --git a/fastNLP/io/data_loader/yelp.py b/fastNLP/io/data_loader/yelp.py deleted file mode 100644 index 333fcab0..00000000 --- a/fastNLP/io/data_loader/yelp.py +++ /dev/null @@ -1,132 +0,0 @@ - -import csv -from typing import Iterable - -from ...core.const import Const -from ...core.dataset import DataSet -from ...core.instance import Instance -from ...core.vocabulary import VocabularyOption, Vocabulary -from ..base_loader import DataBundle, DataSetLoader -from typing import Union, Dict -from ..utils import check_dataloader_paths, get_tokenizer - - -class YelpLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.YelpLoader` :class:`fastNLP.io.data_loader.YelpLoader` - 读取Yelp_full/Yelp_polarity数据集, DataSet包含fields: - - words: list(str), 需要分类的文本 - - target: str, 文本的标签 - - chars:list(str),未index的字符列表 - - 数据集：yelp_full/yelp_polarity - - :param fine_grained: 是否使用SST-5标准，若 ``False`` , 使用SST-2。Default: ``False`` - :param lower: 是否需要自动转小写，默认为False。 - """ - - def __init__(self, fine_grained=False, lower=False): - super(YelpLoader, self).__init__() - tag_v = {'1.0': 'very negative', '2.0': 'negative', '3.0': 'neutral', - '4.0': 'positive', '5.0': 'very positive'} - if not fine_grained: - tag_v['1.0'] = tag_v['2.0'] - tag_v['5.0'] = tag_v['4.0'] - self.fine_grained = fine_grained - self.tag_v = tag_v - self.lower = lower - self.tokenizer = get_tokenizer() - - def _load(self, path): - ds = DataSet() - csv_reader = csv.reader(open(path, encoding='utf-8')) - all_count = 0 - real_count = 0 - for row in csv_reader: - all_count += 1 - if len(row) == 2: - target = self.tag_v[row[0] + ".0"] - words = clean_str(row[1], self.tokenizer, self.lower) - if len(words) != 0: - ds.append(Instance(words=words, target=target)) - real_count += 1 - print("all count:", all_count) - print("real count:", real_count) - return ds - - def process(self, paths: Union[str, Dict[str, str]], - train_ds: Iterable[str] = None, - src_vocab_op: VocabularyOption = None, - tgt_vocab_op: VocabularyOption = None, - char_level_op=False): - paths = check_dataloader_paths(paths) - info = DataBundle(datasets=self.load(paths)) - src_vocab = Vocabulary() if src_vocab_op is None else Vocabulary(**src_vocab_op) - tgt_vocab = Vocabulary(unknown=None, padding=None) \ - if tgt_vocab_op is None else Vocabulary(**tgt_vocab_op) - _train_ds = [info.datasets[name] - for name in train_ds] if train_ds else info.datasets.values() - - def wordtochar(words): - chars = [] - for word in words: - word = word.lower() - for char in word: - chars.append(char) - chars.append('') - chars.pop() - return chars - - input_name, target_name = Const.INPUT, Const.TARGET - info.vocabs = {} - # 就分隔为char形式 - if char_level_op: - for dataset in info.datasets.values(): - dataset.apply_field(wordtochar, field_name=Const.INPUT, new_field_name=Const.CHAR_INPUT) - else: - src_vocab.from_dataset(*_train_ds, field_name=input_name) - src_vocab.index_dataset(*info.datasets.values(), field_name=input_name, new_field_name=input_name) - info.vocabs[input_name] = src_vocab - - tgt_vocab.from_dataset(*_train_ds, field_name=target_name) - tgt_vocab.index_dataset( - *info.datasets.values(), - field_name=target_name, new_field_name=target_name) - - info.vocabs[target_name] = tgt_vocab - - info.datasets['train'], info.datasets['dev'] = info.datasets['train'].split(0.1, shuffle=False) - - for name, dataset in info.datasets.items(): - dataset.set_input(Const.INPUT) - dataset.set_target(Const.TARGET) - - return info - - -def clean_str(sentence, tokenizer, char_lower=False): - """ - heavily borrowed from github - https://github.com/LukeZhuang/Hierarchical-Attention-Network/blob/master/yelp-preprocess.ipynb - :param sentence: is a str - :return: - """ - if char_lower: - sentence = sentence.lower() - import re - nonalpnum = re.compile('[^0-9a-zA-Z?!\']+') - words = tokenizer(sentence) - words_collection = [] - for word in words: - if word in ['-lrb-', '-rrb-', '', '-r', '-l', 'b-']: - continue - tt = nonalpnum.split(word) - t = ''.join(tt) - if t != '': - words_collection.append(t) - - return words_collection - diff --git a/fastNLP/io/dataset_loader.py b/fastNLP/io/dataset_loader.py deleted file mode 100644 index ad6bbdc1..00000000 --- a/fastNLP/io/dataset_loader.py +++ /dev/null @@ -1,138 +0,0 @@ -""" -dataset_loader模块实现了许多 DataSetLoader, 用于读取不同格式的数据, 并返回 `DataSet` , -得到的 :class:`~fastNLP.DataSet` 对象可以直接传入 :class:`~fastNLP.Trainer` 和 :class:`~fastNLP.Tester`, 用于模型的训练和测试。 -以SNLI数据集为例:: - - loader = SNLILoader() - train_ds = loader.load('path/to/train') - dev_ds = loader.load('path/to/dev') - test_ds = loader.load('path/to/test') - - # ... do stuff - -为 fastNLP 提供 DataSetLoader 的开发者请参考 :class:`~fastNLP.io.DataSetLoader` 的介绍。 -""" -__all__ = [ - 'CSVLoader', - 'JsonLoader', -] - - -from ..core.dataset import DataSet -from ..core.instance import Instance -from .file_reader import _read_csv, _read_json -from .base_loader import DataSetLoader - - -class JsonLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.JsonLoader` :class:`fastNLP.io.dataset_loader.JsonLoader` - - 读取json格式数据.数据必须按行存储,每行是一个包含各类属性的json对象 - - :param dict fields: 需要读入的json属性名称, 和读入后在DataSet中存储的field_name - ``fields`` 的 `key` 必须是json对象的属性名. ``fields`` 的 `value` 为读入后在DataSet存储的 `field_name` , - `value` 也可为 ``None`` , 这时读入后的 `field_name` 与json对象对应属性同名 - ``fields`` 可为 ``None`` , 这时,json对象所有属性都保存在DataSet中. Default: ``None`` - :param bool dropna: 是否忽略非法数据,若 ``True`` 则忽略,若 ``False`` ,在遇到非法数据时,抛出 ``ValueError`` . - Default: ``False`` - """ - - def __init__(self, fields=None, dropna=False): - super(JsonLoader, self).__init__() - self.dropna = dropna - self.fields = None - self.fields_list = None - if fields: - self.fields = {} - for k, v in fields.items(): - self.fields[k] = k if v is None else v - self.fields_list = list(self.fields.keys()) - - def _load(self, path): - ds = DataSet() - for idx, d in _read_json(path, fields=self.fields_list, dropna=self.dropna): - if self.fields: - ins = {self.fields[k]: v for k, v in d.items()} - else: - ins = d - ds.append(Instance(**ins)) - return ds - - -class CSVLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.CSVLoader` :class:`fastNLP.io.dataset_loader.CSVLoader` - - 读取CSV格式的数据集。返回 ``DataSet`` - - :param List[str] headers: CSV文件的文件头.定义每一列的属性名称,即返回的DataSet中`field`的名称 - 若为 ``None`` ,则将读入文件的第一行视作 ``headers`` . Default: ``None`` - :param str sep: CSV文件中列与列之间的分隔符. Default: "," - :param bool dropna: 是否忽略非法数据,若 ``True`` 则忽略,若 ``False`` ,在遇到非法数据时,抛出 ``ValueError`` . - Default: ``False`` - """ - - def __init__(self, headers=None, sep=",", dropna=False): - self.headers = headers - self.sep = sep - self.dropna = dropna - - def _load(self, path): - ds = DataSet() - for idx, data in _read_csv(path, headers=self.headers, - sep=self.sep, dropna=self.dropna): - ds.append(Instance(**data)) - return ds - - -def _cut_long_sentence(sent, max_sample_length=200): - """ - 将长于max_sample_length的sentence截成多段，只会在有空格的地方发生截断。 - 所以截取的句子可能长于或者短于max_sample_length - - :param sent: str. - :param max_sample_length: int. - :return: list of str. - """ - sent_no_space = sent.replace(' ', '') - cutted_sentence = [] - if len(sent_no_space) > max_sample_length: - parts = sent.strip().split() - new_line = '' - length = 0 - for part in parts: - length += len(part) - new_line += part + ' ' - if length > max_sample_length: - new_line = new_line[:-1] - cutted_sentence.append(new_line) - length = 0 - new_line = '' - if new_line != '': - cutted_sentence.append(new_line[:-1]) - else: - cutted_sentence.append(sent) - return cutted_sentence - - -def _add_seg_tag(data): - """ - - :param data: list of ([word], [pos], [heads], [head_tags]) - :return: list of ([word], [pos]) - """ - - _processed = [] - for word_list, pos_list, _, _ in data: - new_sample = [] - for word, pos in zip(word_list, pos_list): - if len(word) == 1: - new_sample.append((word, 'S-' + pos)) - else: - new_sample.append((word[0], 'B-' + pos)) - for c in word[1:-1]: - new_sample.append((c, 'M-' + pos)) - new_sample.append((word[-1], 'E-' + pos)) - _processed.append(list(map(list, zip(*new_sample)))) - return _processed diff --git a/fastNLP/io/embed_loader.py b/fastNLP/io/embed_loader.py index 91a0919c..73a7a1de 100644 --- a/fastNLP/io/embed_loader.py +++ b/fastNLP/io/embed_loader.py @@ -1,16 +1,20 @@ +""" +.. todo:: + doc +""" __all__ = [ "EmbedLoader", "EmbeddingOption", ] +import logging import os import warnings import numpy as np -from ..core.vocabulary import Vocabulary -from .base_loader import BaseLoader from ..core.utils import Option +from ..core.vocabulary import Vocabulary class EmbeddingOption(Option): @@ -27,10 +31,8 @@ class EmbeddingOption(Option): ) -class EmbedLoader(BaseLoader): +class EmbedLoader: """ - 别名：:class:`fastNLP.io.EmbedLoader` :class:`fastNLP.io.embed_loader.EmbedLoader` - 用于读取预训练的embedding, 读取结果可直接载入为模型参数。 """ @@ -79,9 +81,9 @@ class EmbedLoader(BaseLoader): word = ''.join(parts[:-dim]) nums = parts[-dim:] # 对齐unk与pad - if word==padding and vocab.padding is not None: + if word == padding and vocab.padding is not None: word = vocab.padding - elif word==unknown and vocab.unknown is not None: + elif word == unknown and vocab.unknown is not None: word = vocab.unknown if word in vocab: index = vocab.to_index(word) @@ -91,10 +93,10 @@ class EmbedLoader(BaseLoader): if error == 'ignore': warnings.warn("Error occurred at the {} line.".format(idx)) else: - print("Error occurred at the {} line.".format(idx)) + logging.error("Error occurred at the {} line.".format(idx)) raise e total_hits = sum(hit_flags) - print("Found {} out of {} words in the pre-training embedding.".format(total_hits, len(vocab))) + logging.info("Found {} out of {} words in the pre-training embedding.".format(total_hits, len(vocab))) if init_method is None: found_vectors = matrix[hit_flags] if len(found_vectors) != 0: @@ -157,7 +159,7 @@ class EmbedLoader(BaseLoader): warnings.warn("Error occurred at the {} line.".format(idx)) pass else: - print("Error occurred at the {} line.".format(idx)) + logging.error("Error occurred at the {} line.".format(idx)) raise e if dim == -1: raise RuntimeError("{} is an empty file.".format(embed_filepath)) @@ -166,7 +168,7 @@ class EmbedLoader(BaseLoader): index = vocab.to_index(key) matrix[index] = vec - if (unknown is not None and not found_unknown) or (padding is not None and not found_pad): + if ((unknown is not None) and (not found_unknown)) or ((padding is not None) and (not found_pad)): start_idx = 0 if padding is not None: start_idx += 1 @@ -175,9 +177,9 @@ class EmbedLoader(BaseLoader): mean = np.mean(matrix[start_idx:], axis=0, keepdims=True) std = np.std(matrix[start_idx:], axis=0, keepdims=True) - if (unknown is not None and not found_unknown): + if (unknown is not None) and (not found_unknown): matrix[start_idx - 1] = np.random.randn(1, dim).astype(dtype) * std + mean - if (padding is not None and not found_pad): + if (padding is not None) and (not found_pad): matrix[0] = np.random.randn(1, dim).astype(dtype) * std + mean if normalize: diff --git a/fastNLP/io/file_reader.py b/fastNLP/io/file_reader.py index 17a0a6ca..b64b115b 100644 --- a/fastNLP/io/file_reader.py +++ b/fastNLP/io/file_reader.py @@ -1,9 +1,14 @@ -""" +"""undocumented 此模块用于给其它模块提供读取文件的函数，没有为用户提供 API """ + +__all__ = [] + import json import csv +from ..core import logger + def _read_csv(path, encoding='utf-8', headers=None, sep=',', dropna=True): """ @@ -17,26 +22,27 @@ def _read_csv(path, encoding='utf-8', headers=None, sep=',', dropna=True): :if False, raise ValueError when reading invalid data. default: True :return: generator, every time yield (line number, csv item) """ - f = csv.reader(open(path, encoding=encoding), delimiter=sep) - start_idx = 0 - if headers is None: - headers = next(f) - start_idx += 1 - elif not isinstance(headers, (list, tuple)): - raise TypeError("headers should be list or tuple, not {}." \ - .format(type(headers))) - for line_idx, line in enumerate(f, start_idx): - contents = line - if len(contents) != len(headers): - if dropna: - continue - else: - raise ValueError("Line {} has {} parts, while header has {} parts." \ - .format(line_idx, len(contents), len(headers))) - _dict = {} - for header, content in zip(headers, contents): - _dict[header] = content - yield line_idx, _dict + with open(path, 'r', encoding=encoding) as csv_file: + f = csv.reader(csv_file, delimiter=sep) + start_idx = 0 + if headers is None: + headers = next(f) + start_idx += 1 + elif not isinstance(headers, (list, tuple)): + raise TypeError("headers should be list or tuple, not {}." \ + .format(type(headers))) + for line_idx, line in enumerate(f, start_idx): + contents = line + if len(contents) != len(headers): + if dropna: + continue + else: + raise ValueError("Line {} has {} parts, while header has {} parts." \ + .format(line_idx, len(contents), len(headers))) + _dict = {} + for header, content in zip(headers, contents): + _dict[header] = content + yield line_idx, _dict def _read_json(path, encoding='utf-8', fields=None, dropna=True): @@ -81,6 +87,7 @@ def _read_conll(path, encoding='utf-8', indexes=None, dropna=True): :if False, raise ValueError when reading invalid data. default: True :return: generator, every time yield (line number, conll item) """ + def parse_conll(sample): sample = list(map(list, zip(*sample))) sample = [sample[i] for i in indexes] @@ -88,14 +95,15 @@ def _read_conll(path, encoding='utf-8', indexes=None, dropna=True): if len(f) <= 0: raise ValueError('empty field') return sample + with open(path, 'r', encoding=encoding) as f: sample = [] start = next(f).strip() - if '-DOCSTART-' not in start and start!='': + if start != '': sample.append(start.split()) for line_idx, line in enumerate(f, 1): line = line.strip() - if line=='': + if line == '': if len(sample): try: res = parse_conll(sample) @@ -103,13 +111,13 @@ def _read_conll(path, encoding='utf-8', indexes=None, dropna=True): yield line_idx, res except Exception as e: if dropna: + logger.warn('Invalid instance which ends at line: {} has been dropped.'.format(line_idx)) continue - raise ValueError('invalid instance ends at line: {}'.format(line_idx)) + raise ValueError('Invalid instance which ends at line: {}'.format(line_idx)) elif line.startswith('#'): continue else: - if not line.startswith('-DOCSTART-'): - sample.append(line.split()) + sample.append(line.split()) if len(sample) > 0: try: res = parse_conll(sample) @@ -117,5 +125,5 @@ def _read_conll(path, encoding='utf-8', indexes=None, dropna=True): except Exception as e: if dropna: return - print('invalid instance ends at line: {}'.format(line_idx)) + logger.error('invalid instance ends at line: {}'.format(line_idx)) raise e diff --git a/fastNLP/io/file_utils.py b/fastNLP/io/file_utils.py index cb762eb7..f76bcd26 100644 --- a/fastNLP/io/file_utils.py +++ b/fastNLP/io/file_utils.py @@ -1,71 +1,159 @@ +""" +.. todo:: + doc +""" + +__all__ = [ + "cached_path", + "get_filepath", + "get_cache_path", + "split_filename_suffix", + "get_from_cache", +] import os +import re +import shutil +import tempfile from pathlib import Path from urllib.parse import urlparse -import re + import requests -import tempfile +from requests import HTTPError from tqdm import tqdm -import shutil -import hashlib +from ..core import logger PRETRAINED_BERT_MODEL_DIR = { - 'en': 'bert-base-cased-f89bfe08.zip', - 'en-base-uncased': 'bert-base-uncased-3413b23c.zip', - 'en-base-cased': 'bert-base-cased-f89bfe08.zip', - 'en-large-uncased': 'bert-large-uncased-20939f45.zip', - 'en-large-cased': 'bert-large-cased-e0cf90fc.zip', - - 'en-large-cased-wwm': 'bert-large-cased-wwm-a457f118.zip', - 'en-large-uncased-wwm': 'bert-large-uncased-wwm-92a50aeb.zip', - 'en-base-cased-mrpc': 'bert-base-cased-finetuned-mrpc-c7099855.zip', - - 'cn': 'bert-base-chinese-29d0a84a.zip', - 'cn-base': 'bert-base-chinese-29d0a84a.zip', - - 'multilingual': 'bert-base-multilingual-cased-1bd364ee.zip', - 'multilingual-base-uncased': 'bert-base-multilingual-uncased-f8730fe4.zip', - 'multilingual-base-cased': 'bert-base-multilingual-cased-1bd364ee.zip', + 'en': 'bert-base-cased.zip', + 'en-large-cased-wwm': 'bert-large-cased-wwm.zip', + 'en-large-uncased-wwm': 'bert-large-uncased-wwm.zip', + + 'en-large-uncased': 'bert-large-uncased.zip', + 'en-large-cased': 'bert-large-cased.zip', + + 'en-base-uncased': 'bert-base-uncased.zip', + 'en-base-cased': 'bert-base-cased.zip', + + 'en-base-cased-mrpc': 'bert-base-cased-finetuned-mrpc.zip', + + 'multi-base-cased': 'bert-base-multilingual-cased.zip', + 'multi-base-uncased': 'bert-base-multilingual-uncased.zip', + + 'cn': 'bert-chinese-wwm.zip', + 'cn-base': 'bert-base-chinese.zip', + 'cn-wwm': 'bert-chinese-wwm.zip', + 'cn-wwm-ext': "bert-chinese-wwm-ext.zip" } PRETRAINED_ELMO_MODEL_DIR = { - 'en': 'elmo_en-d39843fe.tar.gz', - 'cn': 'elmo_cn-5e9b34e2.tar.gz' + 'en': 'elmo_en_Medium.zip', + 'en-small': "elmo_en_Small.zip", + 'en-original-5.5b': 'elmo_en_Original_5.5B.zip', + 'en-original': 'elmo_en_Original.zip', + 'en-medium': 'elmo_en_Medium.zip' } PRETRAIN_STATIC_FILES = { - 'en': 'glove.840B.300d-cc1ad5e1.tar.gz', - 'en-glove-840b-300': 'glove.840B.300d-cc1ad5e1.tar.gz', - 'en-glove-6b-50': "glove.6B.50d-a6028c70.tar.gz", - 'en-word2vec-300': "GoogleNews-vectors-negative300-be166d9d.tar.gz", - 'en-fasttext': "cc.en.300.vec-d53187b2.gz", - 'cn': "tencent_cn-dab24577.tar.gz", - 'cn-fasttext': "cc.zh.300.vec-d68a9bcf.gz", + 'en': 'glove.840B.300d.zip', + + 'en-glove-6b-50d': 'glove.6B.50d.zip', + 'en-glove-6b-100d': 'glove.6B.100d.zip', + 'en-glove-6b-200d': 'glove.6B.200d.zip', + 'en-glove-6b-300d': 'glove.6B.300d.zip', + 'en-glove-42b-300d': 'glove.42B.300d.zip', + 'en-glove-840b-300d': 'glove.840B.300d.zip', + 'en-glove-twitter-27b-25d': 'glove.twitter.27B.25d.zip', + 'en-glove-twitter-27b-50d': 'glove.twitter.27B.50d.zip', + 'en-glove-twitter-27b-100d': 'glove.twitter.27B.100d.zip', + 'en-glove-twitter-27b-200d': 'glove.twitter.27B.200d.zip', + + 'en-word2vec-300': "GoogleNews-vectors-negative300.txt.gz", + + 'en-fasttext-wiki': "wiki-news-300d-1M.vec.zip", + 'en-fasttext-crawl': "crawl-300d-2M.vec.zip", + + 'cn': "tencent_cn.zip", + 'cn-tencent': "tencent_cn.zip", + 'cn-fasttext': "cc.zh.300.vec.gz", + 'cn-sgns-literature-word': 'sgns.literature.word.txt.zip', + 'cn-char-fastnlp-100d': "cn_char_fastnlp_100d.zip", + 'cn-bi-fastnlp-100d': "cn_bi_fastnlp_100d.zip", + "cn-tri-fastnlp-100d": "cn_tri_fastnlp_100d.zip" +} + +DATASET_DIR = { + 'aclImdb': "imdb.zip", + "yelp-review-full": "yelp_review_full.tar.gz", + "yelp-review-polarity": "yelp_review_polarity.tar.gz", + "mnli": "MNLI.zip", + "snli": "SNLI.zip", + "qnli": "QNLI.zip", + "sst-2": "SST-2.zip", + "sst": "SST.zip", + "rte": "RTE.zip", + "msra-ner": "MSRA_NER.zip", + "peopledaily": "peopledaily.zip", + "weibo-ner": "weibo_NER.zip", + + "cws-pku": 'cws_pku.zip', + "cws-cityu": "cws_cityu.zip", + "cws-as": 'cws_as.zip', + "cws-msra": 'cws_msra.zip', + + "chn-senti-corp":"chn_senti_corp.zip" +} + +PRETRAIN_MAP = {'elmo': PRETRAINED_ELMO_MODEL_DIR, + "bert": PRETRAINED_BERT_MODEL_DIR, + "static": PRETRAIN_STATIC_FILES} + +# 用于扩展fastNLP的下载 +FASTNLP_EXTEND_DATASET_URL = 'fastnlp_dataset_url.txt' +FASTNLP_EXTEND_EMBEDDING_URL = {'elmo': 'fastnlp_elmo_url.txt', + 'bert':'fastnlp_bert_url.txt', + 'static': 'fastnlp_static_url.txt' } -def cached_path(url_or_filename: str, cache_dir: Path=None) -> Path: +def cached_path(url_or_filename: str, cache_dir: str = None, name=None) -> Path: """ - 给定一个url或者文件名(可以是具体的文件名，也可以是文件)，先在cache_dir下寻找该文件是否存在，如果不存在则去下载, 并 - 将文件放入到cache_dir中 + 给定一个url，尝试通过url中的解析出来的文件名字filename到{cache_dir}/{name}/{filename}下寻找这个文件， + + 1. 如果cache_dir=None, 则cache_dir=~/.fastNLP/; 否则cache_dir=cache_dir + 2. 如果name=None, 则没有中间的{name}这一层结构；否者中间结构就为{name} + + 如果有该文件，就直接返回路径 + + 如果没有该文件，则尝试用传入的url下载 + + 或者文件名(可以是具体的文件名，也可以是文件夹)，先在cache_dir下寻找该文件是否存在，如果不存在则去下载, 并 + 将文件放入到cache_dir中. + + :param str url_or_filename: 文件的下载url或者文件名称。 + :param str cache_dir: 文件的缓存文件夹。如果为None，将使用"~/.fastNLP"这个默认路径 + :param str name: 中间一层的名称。如embedding, dataset + :return: """ if cache_dir is None: - dataset_cache = Path(get_defalt_path()) + data_cache = Path(get_cache_path()) else: - dataset_cache = cache_dir + data_cache = cache_dir + + if name: + data_cache = os.path.join(data_cache, name) parsed = urlparse(url_or_filename) if parsed.scheme in ("http", "https"): # URL, so get it from the cache (downloading if necessary) - return get_from_cache(url_or_filename, dataset_cache) - elif parsed.scheme == "" and Path(os.path.join(dataset_cache, url_or_filename)).exists(): + return get_from_cache(url_or_filename, Path(data_cache)) + elif parsed.scheme == "" and Path(os.path.join(data_cache, url_or_filename)).exists(): # File, and it exists. - return Path(url_or_filename) + return Path(os.path.join(data_cache, url_or_filename)) elif parsed.scheme == "": # File, but it doesn't exist. - raise FileNotFoundError("file {} not found".format(url_or_filename)) + raise FileNotFoundError("file {} not found in {}.".format(url_or_filename, data_cache)) else: # Something unknown raise ValueError( @@ -75,48 +163,143 @@ def cached_path(url_or_filename: str, cache_dir: Path=None) -> Path: def get_filepath(filepath): """ - 如果filepath中只有一个文件，则直接返回对应的全路径 - :param filepath: + 如果filepath为文件夹， + + 如果内含多个文件, 返回filepath + + 如果只有一个文件, 返回filepath + filename + + 如果filepath为文件 + + 返回filepath + + :param str filepath: 路径 :return: """ if os.path.isdir(filepath): files = os.listdir(filepath) - if len(files)==1: + if len(files) == 1: return os.path.join(filepath, files[0]) else: return filepath - return filepath + elif os.path.isfile(filepath): + return filepath + else: + raise FileNotFoundError(f"{filepath} is not a valid file or directory.") -def get_defalt_path(): +def get_cache_path(): """ - 获取默认的fastNLP存放路径, 如果将FASTNLP_CACHE_PATH设置在了环境变量中，将使用环境变量的值，使得不用每个用户都去下载。 + 获取fastNLP默认cache的存放路径, 如果将FASTNLP_CACHE_PATH设置在了环境变量中，将使用环境变量的值，使得不用每个用户都去下载。 - :return: + :return str: 存放路径 """ if 'FASTNLP_CACHE_DIR' in os.environ: fastnlp_cache_dir = os.environ.get('FASTNLP_CACHE_DIR') - if os.path.exists(fastnlp_cache_dir): + if os.path.isdir(fastnlp_cache_dir): return fastnlp_cache_dir - raise RuntimeError("Some errors happens on cache directory.") - else: - raise RuntimeError("There function is not available right now.") + else: + raise NotADirectoryError(f"{os.environ['FASTNLP_CACHE_DIR']} is not a directory.") fastnlp_cache_dir = os.path.expanduser(os.path.join("~", ".fastNLP")) return fastnlp_cache_dir def _get_base_url(name): + """ + 根据name返回下载的url地址。 + + :param str name: 支持dataset和embedding两种 + :return: + """ # 返回的URL结尾必须是/ - if 'FASTNLP_BASE_URL' in os.environ: - fastnlp_base_url = os.environ['FASTNLP_BASE_URL'] - return fastnlp_base_url - raise RuntimeError("There function is not available right now.") + environ_name = "FASTNLP_{}_URL".format(name.upper()) + + if environ_name in os.environ: + url = os.environ[environ_name] + if url.endswith('/'): + return url + else: + return url + '/' + else: + URLS = { + 'embedding': "http://dbcloud.irocn.cn:8989/api/public/dl/", + "dataset": "http://dbcloud.irocn.cn:8989/api/public/dl/dataset/" + } + if name.lower() not in URLS: + raise KeyError(f"{name} is not recognized.") + return URLS[name.lower()] + + +def _get_embedding_url(embed_type, name): + """ + 给定embedding类似和名称，返回下载url + + :param str embed_type: 支持static, bert, elmo。即embedding的类型 + :param str name: embedding的名称, 例如en, cn, based等 + :return: str, 下载的url地址 + """ + # 从扩展中寻找下载的url + _filename = FASTNLP_EXTEND_EMBEDDING_URL.get(embed_type, None) + if _filename: + url = _read_extend_url_file(_filename, name) + if url: + return url + embed_map = PRETRAIN_MAP.get(embed_type, None) + if embed_map: + filename = embed_map.get(name, None) + if filename: + url = _get_base_url('embedding') + filename + return url + raise KeyError("There is no {}. Only supports {}.".format(name, list(embed_map.keys()))) + else: + raise KeyError(f"There is no {embed_type}. Only supports bert, elmo, static") + +def _read_extend_url_file(filename, name)->str: + """ + filename中的内容使用制表符隔开，第一列是名称，第二列是下载的url地址 + + :param str filename: 在默认的路径下寻找file这个文件 + :param str name: 需要寻找的资源的名称 + :return: str or None + """ + cache_dir = get_cache_path() + filepath = os.path.join(cache_dir, filename) + if os.path.exists(filepath): + with open(filepath, 'r', encoding='utf-8') as f: + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + if len(parts) == 2: + if name == parts[0]: + return parts[1] + return None + +def _get_dataset_url(name): + """ + 给定dataset的名称，返回下载url + + :param str name: 给定dataset的名称，比如imdb, sst-2等 + :return: str + """ + # 从扩展中寻找下载的url + url = _read_extend_url_file(FASTNLP_EXTEND_DATASET_URL, name) + if url: + return url + + filename = DATASET_DIR.get(name, None) + if filename: + url = _get_base_url('dataset') + filename + return url + else: + raise KeyError(f"There is no {name}.") def split_filename_suffix(filepath): """ - 给定filepath返回对应的name和suffix - :param filepath: + 给定filepath 返回对应的name和suffix. 如果后缀是多个点，仅支持.tar.gz类型 + + :param filepath: 文件路径 :return: filename, suffix """ filename = os.path.basename(filepath) @@ -127,21 +310,19 @@ def split_filename_suffix(filepath): def get_from_cache(url: str, cache_dir: Path = None) -> Path: """ - 尝试在cache_dir中寻找url定义的资源; 如果没有找到。则从url下载并将结果放在cache_dir下，缓存的名称由url的结果推断而来。 - 如果从url中下载的资源解压后有多个文件，则返回directory的路径; 如果只有一个资源，则返回具体的路径。 - + 尝试在cache_dir中寻找url定义的资源; 如果没有找到; 则从url下载并将结果放在cache_dir下，缓存的名称由url的结果推断而来。会将下载的 + 文件解压，将解压后的文件全部放在cache_dir文件夹中。 + + 如果从url中下载的资源解压后有多个文件，则返回目录的路径; 如果只有一个资源文件，则返回具体的路径。 + + :param url: 资源的 url + :param cache_dir: cache 目录 + :return: 路径 """ cache_dir.mkdir(parents=True, exist_ok=True) filename = re.sub(r".+/", "", url) dir_name, suffix = split_filename_suffix(filename) - sep_index = dir_name[::-1].index('-') - if sep_index<0: - check_sum = None - else: - check_sum = dir_name[-sep_index+1:] - sep_index = len(dir_name) if sep_index==-1 else -sep_index-1 - dir_name = dir_name[:sep_index] # 寻找与它名字匹配的内容, 而不关心后缀 match_dir_name = match_file(dir_name, cache_dir) @@ -154,11 +335,11 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: return get_filepath(cache_path) # make HEAD request to check ETag TODO ETag可以用来判断资源是否已经更新了，之后需要加上 - response = requests.head(url, headers={"User-Agent": "fastNLP"}) - if response.status_code != 200: - raise IOError( - f"HEAD request failed for url {url} with status code {response.status_code}." - ) + # response = requests.head(url, headers={"User-Agent": "fastNLP"}) + # if response.status_code != 200: + # raise IOError( + # f"HEAD request failed for url {url} with status code {response.status_code}." + # ) # add ETag to filename if it exists # etag = response.headers.get("ETag") @@ -166,74 +347,77 @@ def get_from_cache(url: str, cache_dir: Path = None) -> Path: if not cache_path.exists(): # Download to temporary file, then copy to cache dir once finished. # Otherwise you get corrupt cache entries if the download gets interrupted. - fd, temp_filename = tempfile.mkstemp() - print("%s not found in cache, downloading to %s"%(url, temp_filename)) - # GET file object req = requests.get(url, stream=True, headers={"User-Agent": "fastNLP"}) - content_length = req.headers.get("Content-Length") - total = int(content_length) if content_length is not None else None - progress = tqdm(unit="B", total=total) - sha256 = hashlib.sha256() - with open(temp_filename, "wb") as temp_file: - for chunk in req.iter_content(chunk_size=1024): - if chunk: # filter out keep-alive new chunks - progress.update(len(chunk)) - temp_file.write(chunk) - sha256.update(chunk) - # check sum - digit = sha256.hexdigest()[:8] - if not check_sum: - assert digit == check_sum, "File corrupted when download." - progress.close() - print(f"Finish download from {url}.") - - # 开始解压 - delete_temp_dir = None - if suffix in ('.zip', '.tar.gz'): - uncompress_temp_dir = tempfile.mkdtemp() - delete_temp_dir = uncompress_temp_dir - print(f"Start to uncompress file to {uncompress_temp_dir}.") - if suffix == '.zip': - unzip_file(Path(temp_filename), Path(uncompress_temp_dir)) - else: - untar_gz_file(Path(temp_filename), Path(uncompress_temp_dir)) - filenames = os.listdir(uncompress_temp_dir) - if len(filenames)==1: - if os.path.isdir(os.path.join(uncompress_temp_dir, filenames[0])): - uncompress_temp_dir = os.path.join(uncompress_temp_dir, filenames[0]) - - cache_path.mkdir(parents=True, exist_ok=True) - print("Finish un-compressing file.") - else: - uncompress_temp_dir = temp_filename - cache_path = str(cache_path) + suffix - success = False - try: - # 复制到指定的位置 - print(f"Copy file to {cache_path}.") - if os.path.isdir(uncompress_temp_dir): - for filename in os.listdir(uncompress_temp_dir): - shutil.copyfile(os.path.join(uncompress_temp_dir, filename), cache_path/filename) - else: - shutil.copyfile(uncompress_temp_dir, cache_path) - success = True - except Exception as e: - print(e) - raise e - finally: - if not success: - if cache_path.exists(): - if cache_path.is_file(): - os.remove(cache_path) + if req.status_code == 200: + success = False + fd, temp_filename = tempfile.mkstemp() + uncompress_temp_dir = None + try: + content_length = req.headers.get("Content-Length") + total = int(content_length) if content_length is not None else None + progress = tqdm(unit="B", total=total, unit_scale=1) + logger.info("%s not found in cache, downloading to %s" % (url, temp_filename)) + + with open(temp_filename, "wb") as temp_file: + for chunk in req.iter_content(chunk_size=1024 * 16): + if chunk: # filter out keep-alive new chunks + progress.update(len(chunk)) + temp_file.write(chunk) + progress.close() + logger.info(f"Finish download from {url}") + + # 开始解压 + if suffix in ('.zip', '.tar.gz', '.gz'): + uncompress_temp_dir = tempfile.mkdtemp() + logger.debug(f"Start to uncompress file to {uncompress_temp_dir}") + if suffix == '.zip': + unzip_file(Path(temp_filename), Path(uncompress_temp_dir)) + elif suffix == '.gz': + ungzip_file(temp_filename, uncompress_temp_dir, dir_name) else: - shutil.rmtree(cache_path) - if delete_temp_dir: - shutil.rmtree(delete_temp_dir) - os.close(fd) - os.remove(temp_filename) - - return get_filepath(cache_path) + untar_gz_file(Path(temp_filename), Path(uncompress_temp_dir)) + filenames = os.listdir(uncompress_temp_dir) + if len(filenames) == 1: + if os.path.isdir(os.path.join(uncompress_temp_dir, filenames[0])): + uncompress_temp_dir = os.path.join(uncompress_temp_dir, filenames[0]) + + cache_path.mkdir(parents=True, exist_ok=True) + logger.debug("Finish un-compressing file.") + else: + uncompress_temp_dir = temp_filename + cache_path = str(cache_path) + suffix + + # 复制到指定的位置 + logger.info(f"Copy file to {cache_path}") + if os.path.isdir(uncompress_temp_dir): + for filename in os.listdir(uncompress_temp_dir): + if os.path.isdir(os.path.join(uncompress_temp_dir, filename)): + shutil.copytree(os.path.join(uncompress_temp_dir, filename), cache_path / filename) + else: + shutil.copyfile(os.path.join(uncompress_temp_dir, filename), cache_path / filename) + else: + shutil.copyfile(uncompress_temp_dir, cache_path) + success = True + except Exception as e: + logger.error(e) + raise e + finally: + if not success: + if cache_path.exists(): + if cache_path.is_file(): + os.remove(cache_path) + else: + shutil.rmtree(cache_path) + os.close(fd) + os.remove(temp_filename) + if os.path.isdir(uncompress_temp_dir): + shutil.rmtree(uncompress_temp_dir) + elif os.path.isfile(uncompress_temp_dir): + os.remove(uncompress_temp_dir) + return get_filepath(cache_path) + else: + raise HTTPError(f"Status code:{req.status_code}. Fail to download from {url}.") def unzip_file(file: Path, to: Path): @@ -245,55 +429,39 @@ def unzip_file(file: Path, to: Path): zipObj.extractall(to) -def untar_gz_file(file:Path, to:Path): +def untar_gz_file(file: Path, to: Path): import tarfile with tarfile.open(file, 'r:gz') as tar: tar.extractall(to) -def match_file(dir_name: str, cache_dir: str) -> str: +def ungzip_file(file: str, to: str, filename:str): + import gzip + + g_file = gzip.GzipFile(file) + with open(os.path.join(to, filename), 'wb+') as f: + f.write(g_file.read()) + g_file.close() + + +def match_file(dir_name: str, cache_dir: Path) -> str: """ - 匹配的原则是，在cache_dir下的文件: (1) 与dir_name完全一致; (2) 除了后缀以外和dir_name完全一致。 + 匹配的原则是: 在cache_dir下的文件与dir_name完全一致, 或除了后缀以外和dir_name完全一致。 如果找到了两个匹配的结果将报错. 如果找到了则返回匹配的文件的名称; 没有找到返回空字符串 :param dir_name: 需要匹配的名称 :param cache_dir: 在该目录下找匹配dir_name是否存在 - :return: str + :return str: 做为匹配结果的字符串 """ files = os.listdir(cache_dir) matched_filenames = [] for file_name in files: - if re.match(dir_name+'$', file_name) or re.match(dir_name+'\\..*', file_name): + if re.match(dir_name + '$', file_name) or re.match(dir_name + '\\..*', file_name): matched_filenames.append(file_name) - if len(matched_filenames)==0: + if len(matched_filenames) == 0: return '' - elif len(matched_filenames)==1: + elif len(matched_filenames) == 1: return matched_filenames[-1] else: raise RuntimeError(f"Duplicate matched files:{matched_filenames}, this should be caused by a bug.") - - -if __name__ == '__main__': - cache_dir = Path('caches') - cache_dir = None - # 需要对cache_dir进行测试 - base_url = 'http://0.0.0.0:8888/file/download' - # if True: - # for filename in os.listdir(cache_dir): - # if os.path.isdir(os.path.join(cache_dir, filename)): - # shutil.rmtree(os.path.join(cache_dir, filename)) - # else: - # os.remove(os.path.join(cache_dir, filename)) - # 1. 测试.txt文件 - print(cached_path(base_url + '/{}'.format('txt_test-bcb4fe65.txt'), cache_dir)) - # 2. 测试.zip文件(只有一个文件) - print(cached_path(base_url + '/{}'.format('zip_test-40966d39.zip'), cache_dir)) - # 3. 测试.zip文件(有多个文件) - print(cached_path(base_url + '/{}'.format('zip_pack_test-70c0b20d.zip'), cache_dir)) - # 4. 测试.tar.gz文件 - print(cached_path(base_url + '/{}'.format('tar_gz_test-3e2679cf.tar.gz'), cache_dir)) - # 5. 测试.tar.gz多个文件 - print(cached_path(base_url + '/{}'.format('tar_gz_pack_test-08dfdccd.tar.gz'), cache_dir)) - - # 6. 测试.pkl文件 diff --git a/fastNLP/io/loader/__init__.py b/fastNLP/io/loader/__init__.py new file mode 100644 index 00000000..3ad1b47d --- /dev/null +++ b/fastNLP/io/loader/__init__.py @@ -0,0 +1,84 @@ +""" +Loader用于读取数据，并将内容读取到 :class:`~fastNLP.DataSet` 或者 :class:`~fastNLP.io.DataBundle` 中。所有的Loader都支持以下的 +三个方法： ``__init__`` ， ``_load`` , ``loads`` . 其中 ``__init__(...)`` 用于申明读取参数，以及说明该Loader支持的数据格式， +读取后 :class:`~fastNLP.DataSet` 中的 `field` ; ``_load(path)`` 方法传入文件路径读取单个文件，并返回 :class:`~fastNLP.DataSet` ; +``load(paths)`` 用于读取文件夹下的文件，并返回 :class:`~fastNLP.io.DataBundle` 类型的对象 , load()方法支持以下几种类型的参数: + +0.传入None + 将尝试自动下载数据集并缓存。但不是所有的数据都可以直接下载。 + +1.传入一个文件的 path + 返回的 `data_bundle` 包含一个名为 `train` 的 dataset ,可以通过 ``data_bundle.datasets['train']`` 获取 + +2.传入一个文件夹目录 + 将读取的是这个文件夹下文件名中包含 `train` , `test` , `dev` 的文件，其它文件会被忽略。假设某个目录下的文件为:: + + | + +-train.txt + +-dev.txt + +-test.txt + +-other.txt + + 在 Loader().load('/path/to/dir') 返回的 `data_bundle` 中可以用 ``data_bundle.datasets['train']`` , ``data_bundle.datasets['dev']`` , + ``data_bundle.datasets['test']`` 获取对应的 `dataset` ，其中 `other.txt` 的内容会被忽略。假设某个目录下的文件为:: + + | + +-train.txt + +-dev.txt + + 在 Loader().load('/path/to/dir') 返回的 `data_bundle` 中可以用 ``data_bundle.datasets['train']`` , + ``data_bundle.datasets['dev']`` 获取对应的 dataset。 + +3.传入一个字典 + 字典的的 key 为 `dataset` 的名称，value 是该 `dataset` 的文件路径:: + + paths = {'train':'/path/to/train', 'dev': '/path/to/dev', 'test':'/path/to/test'} + + 在 Loader().load(paths) 返回的 `data_bundle` 中可以用 ``data_bundle.datasets['train']`` , ``data_bundle.datasets['dev']`` , + ``data_bundle.datasets['test']`` 来获取对应的 `dataset` + +fastNLP 目前提供了如下的 Loader + + + +""" + +__all__ = [ + 'Loader', + + 'YelpLoader', + 'YelpFullLoader', + 'YelpPolarityLoader', + 'IMDBLoader', + 'SSTLoader', + 'SST2Loader', + "ChnSentiCorpLoader", + + 'ConllLoader', + 'Conll2003Loader', + 'Conll2003NERLoader', + 'OntoNotesNERLoader', + 'CTBLoader', + "MsraNERLoader", + "PeopleDailyNERLoader", + "WeiboNERLoader", + + 'CSVLoader', + 'JsonLoader', + + 'CWSLoader', + + 'MNLILoader', + "QuoraLoader", + "SNLILoader", + "QNLILoader", + "RTELoader" +] +from .classification import YelpLoader, YelpFullLoader, YelpPolarityLoader, IMDBLoader, SSTLoader, SST2Loader, ChnSentiCorpLoader +from .conll import ConllLoader, Conll2003Loader, Conll2003NERLoader, OntoNotesNERLoader, CTBLoader +from .csv import CSVLoader +from .cws import CWSLoader +from .json import JsonLoader +from .loader import Loader +from .matching import MNLILoader, QuoraLoader, SNLILoader, QNLILoader, RTELoader +from .conll import MsraNERLoader, PeopleDailyNERLoader, WeiboNERLoader diff --git a/fastNLP/io/loader/classification.py b/fastNLP/io/loader/classification.py new file mode 100644 index 00000000..9efcf5d2 --- /dev/null +++ b/fastNLP/io/loader/classification.py @@ -0,0 +1,399 @@ +"""undocumented""" + +__all__ = [ + "YelpLoader", + "YelpFullLoader", + "YelpPolarityLoader", + "IMDBLoader", + "SSTLoader", + "SST2Loader", + "ChnSentiCorpLoader" +] + +import glob +import os +import random +import shutil +import time +import warnings + +from .loader import Loader +from ...core.dataset import DataSet +from ...core.instance import Instance + + +class YelpLoader(Loader): + """ + 原始数据中内容应该为, 每一行为一个sample，第一个逗号之前为target，第一个逗号之后为文本内容。 + + Example:: + + "1","I got 'new' tires from the..." + "1","Don't waste your time..." + + 读取YelpFull, YelpPolarity的数据。可以通过xxx下载并预处理数据。 + 读取的DataSet将具备以下的数据结构 + + .. csv-table:: + :header: "raw_words", "target" + + "I got 'new' tires from them and... ", "1" + "Don't waste your time. We had two...", "1" + "...", "..." + + """ + + def __init__(self): + super(YelpLoader, self).__init__() + + def _load(self, path: str = None): + ds = DataSet() + with open(path, 'r', encoding='utf-8') as f: + for line in f: + line = line.strip() + sep_index = line.index(',') + target = line[:sep_index] + raw_words = line[sep_index + 1:] + if target.startswith("\""): + target = target[1:] + if target.endswith("\""): + target = target[:-1] + if raw_words.endswith("\""): + raw_words = raw_words[:-1] + if raw_words.startswith('"'): + raw_words = raw_words[1:] + raw_words = raw_words.replace('""', '"') # 替换双引号 + if raw_words: + ds.append(Instance(raw_words=raw_words, target=target)) + return ds + + +class YelpFullLoader(YelpLoader): + def download(self, dev_ratio: float = 0.1, re_download: bool = False): + """ + 自动下载数据集，如果你使用了这个数据集，请引用以下的文章 + + Xiang Zhang, Junbo Zhao, Yann LeCun. Character-level Convolutional Networks for Text Classification. Advances + in Neural Information Processing Systems 28 (NIPS 2015) + + 根据dev_ratio的值随机将train中的数据取出一部分作为dev数据。下载完成后在output_dir中有train.csv, test.csv, + dev.csv三个文件。 + + :param float dev_ratio: 如果路径中没有dev集，从train划分多少作为dev的数据. 如果为0，则不划分dev。 + :param bool re_download: 是否重新下载数据，以重新切分数据。 + :return: str, 数据集的目录地址 + """ + + dataset_name = 'yelp-review-full' + data_dir = self._get_dataset_path(dataset_name=dataset_name) + modify_time = 0 + for filepath in glob.glob(os.path.join(data_dir, '*')): + modify_time = os.stat(filepath).st_mtime + break + if time.time() - modify_time > 1 and re_download: # 通过这种比较丑陋的方式判断一下文件是否是才下载的 + shutil.rmtree(data_dir) + data_dir = self._get_dataset_path(dataset_name=dataset_name) + + if not os.path.exists(os.path.join(data_dir, 'dev.csv')): + if dev_ratio > 0: + assert 0 < dev_ratio < 1, "dev_ratio should be in range (0,1)." + try: + with open(os.path.join(data_dir, 'train.csv'), 'r', encoding='utf-8') as f, \ + open(os.path.join(data_dir, 'middle_file.csv'), 'w', encoding='utf-8') as f1, \ + open(os.path.join(data_dir, 'dev.csv'), 'w', encoding='utf-8') as f2: + for line in f: + if random.random() < dev_ratio: + f2.write(line) + else: + f1.write(line) + os.remove(os.path.join(data_dir, 'train.csv')) + os.renames(os.path.join(data_dir, 'middle_file.csv'), os.path.join(data_dir, 'train.csv')) + finally: + if os.path.exists(os.path.join(data_dir, 'middle_file.csv')): + os.remove(os.path.join(data_dir, 'middle_file.csv')) + + return data_dir + + +class YelpPolarityLoader(YelpLoader): + def download(self, dev_ratio: float = 0.1, re_download=False): + """ + 自动下载数据集，如果你使用了这个数据集，请引用以下的文章 + + Xiang Zhang, Junbo Zhao, Yann LeCun. Character-level Convolutional Networks for Text Classification. Advances + in Neural Information Processing Systems 28 (NIPS 2015) + + 根据dev_ratio的值随机将train中的数据取出一部分作为dev数据。下载完成后从train中切分dev_ratio这么多作为dev + + :param float dev_ratio: 如果路径中不存在dev.csv, 从train划分多少作为dev的数据。 如果为0，则不划分dev。 + :param bool re_download: 是否重新下载数据，以重新切分数据。 + :return: str, 数据集的目录地址 + """ + dataset_name = 'yelp-review-polarity' + data_dir = self._get_dataset_path(dataset_name=dataset_name) + modify_time = 0 + for filepath in glob.glob(os.path.join(data_dir, '*')): + modify_time = os.stat(filepath).st_mtime + break + if time.time() - modify_time > 1 and re_download: # 通过这种比较丑陋的方式判断一下文件是否是才下载的 + shutil.rmtree(data_dir) + data_dir = self._get_dataset_path(dataset_name=dataset_name) + + if not os.path.exists(os.path.join(data_dir, 'dev.csv')): + if dev_ratio > 0: + assert 0 < dev_ratio < 1, "dev_ratio should be in range (0,1)." + try: + with open(os.path.join(data_dir, 'train.csv'), 'r', encoding='utf-8') as f, \ + open(os.path.join(data_dir, 'middle_file.csv'), 'w', encoding='utf-8') as f1, \ + open(os.path.join(data_dir, 'dev.csv'), 'w', encoding='utf-8') as f2: + for line in f: + if random.random() < dev_ratio: + f2.write(line) + else: + f1.write(line) + os.remove(os.path.join(data_dir, 'train.csv')) + os.renames(os.path.join(data_dir, 'middle_file.csv'), os.path.join(data_dir, 'train.csv')) + finally: + if os.path.exists(os.path.join(data_dir, 'middle_file.csv')): + os.remove(os.path.join(data_dir, 'middle_file.csv')) + + return data_dir + + +class IMDBLoader(Loader): + """ + IMDBLoader读取后的数据将具有以下两列内容: raw_words: str, 需要分类的文本; target: str, 文本的标签 + DataSet具备以下的结构: + + .. csv-table:: + :header: "raw_words", "target" + + "Bromwell High is a cartoon ... ", "pos" + "Story of a man who has ...", "neg" + "...", "..." + + """ + + def __init__(self): + super(IMDBLoader, self).__init__() + + def _load(self, path: str): + dataset = DataSet() + with open(path, 'r', encoding="utf-8") as f: + for line in f: + line = line.strip() + if not line: + continue + parts = line.split('\t') + target = parts[0] + words = parts[1] + if words: + dataset.append(Instance(raw_words=words, target=target)) + + if len(dataset) == 0: + raise RuntimeError(f"{path} has no valid data.") + + return dataset + + def download(self, dev_ratio: float = 0.1, re_download=False): + """ + 自动下载数据集，如果你使用了这个数据集，请引用以下的文章 + + http://www.aclweb.org/anthology/P11-1015 + + 根据dev_ratio的值随机将train中的数据取出一部分作为dev数据。下载完成后从train中切分0.1作为dev + + :param float dev_ratio: 如果路径中没有dev.txt。从train划分多少作为dev的数据. 如果为0，则不划分dev + :param bool re_download: 是否重新下载数据，以重新切分数据。 + :return: str, 数据集的目录地址 + """ + dataset_name = 'aclImdb' + data_dir = self._get_dataset_path(dataset_name=dataset_name) + modify_time = 0 + for filepath in glob.glob(os.path.join(data_dir, '*')): + modify_time = os.stat(filepath).st_mtime + break + if time.time() - modify_time > 1 and re_download: # 通过这种比较丑陋的方式判断一下文件是否是才下载的 + shutil.rmtree(data_dir) + data_dir = self._get_dataset_path(dataset_name=dataset_name) + + if not os.path.exists(os.path.join(data_dir, 'dev.csv')): + if dev_ratio > 0: + assert 0 < dev_ratio < 1, "dev_ratio should be in range (0,1)." + try: + with open(os.path.join(data_dir, 'train.txt'), 'r', encoding='utf-8') as f, \ + open(os.path.join(data_dir, 'middle_file.txt'), 'w', encoding='utf-8') as f1, \ + open(os.path.join(data_dir, 'dev.txt'), 'w', encoding='utf-8') as f2: + for line in f: + if random.random() < dev_ratio: + f2.write(line) + else: + f1.write(line) + os.remove(os.path.join(data_dir, 'train.txt')) + os.renames(os.path.join(data_dir, 'middle_file.txt'), os.path.join(data_dir, 'train.txt')) + finally: + if os.path.exists(os.path.join(data_dir, 'middle_file.txt')): + os.remove(os.path.join(data_dir, 'middle_file.txt')) + + return data_dir + + +class SSTLoader(Loader): + """ + 读取之后的DataSet具有以下的结构 + + .. csv-table:: 下面是使用SSTLoader读取的DataSet所具备的field + :header: "raw_words" + + "(3 (2 It) (4 (4 (2 's) (4 (3 (2 a)..." + "(4 (4 (2 Offers) (3 (3 (2 that) (3 (3 rare)..." + "..." + + raw_words列是str。 + + """ + + def __init__(self): + super().__init__() + + def _load(self, path: str): + """ + 从path读取SST文件 + + :param str path: 文件路径 + :return: DataSet + """ + ds = DataSet() + with open(path, 'r', encoding='utf-8') as f: + for line in f: + line = line.strip() + if line: + ds.append(Instance(raw_words=line)) + return ds + + def download(self): + """ + 自动下载数据集，如果你使用了这个数据集，请引用以下的文章 + + https://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf + + :return: str, 数据集的目录地址 + """ + output_dir = self._get_dataset_path(dataset_name='sst') + return output_dir + + +class SST2Loader(Loader): + """ + 数据SST2的Loader + 读取之后DataSet将如下所示 + + .. csv-table:: 下面是使用SSTLoader读取的DataSet所具备的field + :header: "raw_words", "target" + + "it 's a charming and often affecting...", "1" + "unflinchingly bleak and...", "0" + "..." + + test的DataSet没有target列。 + """ + + def __init__(self): + super().__init__() + + def _load(self, path: str): + """ + 从path读取SST2文件 + + :param str path: 数据路径 + :return: DataSet + """ + ds = DataSet() + + with open(path, 'r', encoding='utf-8') as f: + f.readline() # 跳过header + if 'test' in os.path.split(path)[1]: + warnings.warn("SST2's test file has no target.") + for line in f: + line = line.strip() + if line: + sep_index = line.index('\t') + raw_words = line[sep_index + 1:] + if raw_words: + ds.append(Instance(raw_words=raw_words)) + else: + for line in f: + line = line.strip() + if line: + raw_words = line[:-2] + target = line[-1] + if raw_words: + ds.append(Instance(raw_words=raw_words, target=target)) + return ds + + def download(self): + """ + 自动下载数据集，如果你使用了该数据集，请引用以下的文章 + + https://nlp.stanford.edu/pubs/SocherBauerManningNg_ACL2013.pdf + + :return: + """ + output_dir = self._get_dataset_path(dataset_name='sst-2') + return output_dir + + +class ChnSentiCorpLoader(Loader): + """ + 支持读取的数据的格式为，第一行为标题(具体内容会被忽略)，之后一行为一个sample，第一个制表符之前被认为是label，第 + 一个制表符及之后认为是句子 + + Example:: + + label raw_chars + 1 這間酒店環境和服務態度亦算不錯,但房間空間太小~~ + 1 <荐书> 推荐所有喜欢<红楼>的红迷们一定要收藏这本书,要知道... + 0 商品的不足暂时还没发现，京东的订单处理速度实在.......周二就打包完成，周五才发货... + + 读取后的DataSet具有以下的field + + .. csv-table:: + :header: "raw_chars", "target" + + "這間酒店環境和服務態度亦算不錯,但房間空間太小~~", "1" + "<荐书> 推荐所有喜欢<红楼>...", "1" + "..." + + """ + def __init__(self): + super().__init__() + + def _load(self, path:str): + """ + 从path中读取数据 + + :param path: + :return: + """ + ds = DataSet() + with open(path, 'r', encoding='utf-8') as f: + f.readline() + for line in f: + line = line.strip() + tab_index = line.index('\t') + if tab_index!=-1: + target = line[:tab_index] + raw_chars = line[tab_index+1:] + if raw_chars: + ds.append(Instance(raw_chars=raw_chars, target=target)) + return ds + + def download(self)->str: + """ + 自动下载数据，该数据取自https://github.com/pengming617/bert_classification/tree/master/data，在 + https://arxiv.org/pdf/1904.09223.pdf与https://arxiv.org/pdf/1906.08101.pdf有使用 + + :return: + """ + output_dir = self._get_dataset_path('chn-senti-corp') + return output_dir diff --git a/fastNLP/io/loader/conll.py b/fastNLP/io/loader/conll.py new file mode 100644 index 00000000..f30b031f --- /dev/null +++ b/fastNLP/io/loader/conll.py @@ -0,0 +1,453 @@ +"""undocumented""" + +__all__ = [ + "ConllLoader", + "Conll2003Loader", + "Conll2003NERLoader", + "OntoNotesNERLoader", + "CTBLoader", + "CNNERLoader", + "MsraNERLoader", + "WeiboNERLoader", + "PeopleDailyNERLoader" +] + +import glob +import os +import random +import shutil +import time + +from .loader import Loader +from ..file_reader import _read_conll +from ...core.const import Const +from ...core.dataset import DataSet +from ...core.instance import Instance + + +class ConllLoader(Loader): + """ + ConllLoader支持读取的数据格式: 以空行隔开两个sample，除了分割行，每一行用空格或者制表符隔开不同的元素。如下例所示: + + Example:: + + # 文件中的内容 + Nadim NNP B-NP B-PER + Ladki NNP I-NP I-PER + + AL-AIN NNP B-NP B-LOC + United NNP B-NP B-LOC + Arab NNP I-NP I-LOC + Emirates NNPS I-NP I-LOC + 1996-12-06 CD I-NP O + ... + + # 如果用以下的参数读取，返回的DataSet将包含raw_words和pos两个field, 这两个field的值分别取自于第0列与第1列 + dataset = ConllLoader(headers=['raw_words', 'pos'], indexes=[0, 1])._load('/path/to/train.conll') + # 如果用以下的参数读取，返回的DataSet将包含raw_words和ner两个field, 这两个field的值分别取自于第0列与第2列 + dataset = ConllLoader(headers=['raw_words', 'ner'], indexes=[0, 3])._load('/path/to/train.conll') + # 如果用以下的参数读取，返回的DataSet将包含raw_words, pos和ner三个field + dataset = ConllLoader(headers=['raw_words', 'pos', 'ner'], indexes=[0, 1, 3])._load('/path/to/train.conll') + + ConllLoader返回的DataSet的field由传入的headers确定。 + + 数据中以"-DOCSTART-"开头的行将被忽略，因为该符号在conll 2003中被用为文档分割符。 + + :param list headers: 每一列数据的名称，需为List or Tuple of str。``header`` 与 ``indexes`` 一一对应 + :param list indexes: 需要保留的数据列下标，从0开始。若为 ``None`` ，则所有列都保留。Default: ``None`` + :param bool dropna: 是否忽略非法数据，若 ``False`` ，遇到非法数据时抛出 ``ValueError`` 。Default: ``True`` + + """ + + def __init__(self, headers, indexes=None, dropna=True): + super(ConllLoader, self).__init__() + if not isinstance(headers, (list, tuple)): + raise TypeError( + 'invalid headers: {}, should be list of strings'.format(headers)) + self.headers = headers + self.dropna = dropna + if indexes is None: + self.indexes = list(range(len(self.headers))) + else: + if len(indexes) != len(headers): + raise ValueError + self.indexes = indexes + + def _load(self, path): + """ + 传入的一个文件路径，将该文件读入DataSet中，field由ConllLoader初始化时指定的headers决定。 + + :param str path: 文件的路径 + :return: DataSet + """ + ds = DataSet() + for idx, data in _read_conll(path, indexes=self.indexes, dropna=self.dropna): + ins = {h: data[i] for i, h in enumerate(self.headers)} + ds.append(Instance(**ins)) + return ds + + +class Conll2003Loader(ConllLoader): + """ + 用于读取conll2003任务的数据。数据的内容应该类似与以下的内容, 第一列为raw_words, 第二列为pos, 第三列为chunking，第四列为ner。 + + Example:: + + Nadim NNP B-NP B-PER + Ladki NNP I-NP I-PER + + AL-AIN NNP B-NP B-LOC + United NNP B-NP B-LOC + Arab NNP I-NP I-LOC + Emirates NNPS I-NP I-LOC + 1996-12-06 CD I-NP O + ... + + 返回的DataSet的内容为 + + .. csv-table:: 下面是Conll2003Loader加载后数据具备的结构。 + :header: "raw_words", "pos", "chunk", "ner" + + "[Nadim, Ladki]", "[NNP, NNP]", "[B-NP, I-NP]", "[B-PER, I-PER]" + "[AL-AIN, United, Arab, ...]", "[NNP, NNP, NNP, ...]", "[B-NP, B-NP, I-NP, ...]", "[B-LOC, B-LOC, I-LOC, ...]" + "[...]", "[...]", "[...]", "[...]" + + """ + + def __init__(self): + headers = [ + 'raw_words', 'pos', 'chunk', 'ner', + ] + super(Conll2003Loader, self).__init__(headers=headers) + + def _load(self, path): + """ + 传入的一个文件路径，将该文件读入DataSet中，field由ConllLoader初始化时指定的headers决定。 + + :param str path: 文件的路径 + :return: DataSet + """ + ds = DataSet() + for idx, data in _read_conll(path, indexes=self.indexes, dropna=self.dropna): + doc_start = False + for i, h in enumerate(self.headers): + field = data[i] + if str(field[0]).startswith('-DOCSTART-'): + doc_start = True + break + if doc_start: + continue + ins = {h: data[i] for i, h in enumerate(self.headers)} + ds.append(Instance(**ins)) + return ds + + def download(self, output_dir=None): + raise RuntimeError("conll2003 cannot be downloaded automatically.") + + +class Conll2003NERLoader(ConllLoader): + """ + 用于读取conll2003任务的NER数据。 + + Example:: + + Nadim NNP B-NP B-PER + Ladki NNP I-NP I-PER + + AL-AIN NNP B-NP B-LOC + United NNP B-NP B-LOC + Arab NNP I-NP I-LOC + Emirates NNPS I-NP I-LOC + 1996-12-06 CD I-NP O + ... + + 返回的DataSet的内容为 + + .. csv-table:: 下面是Conll2003Loader加载后数据具备的结构, target是BIO2编码 + :header: "raw_words", "target" + + "[Nadim, Ladki]", "[B-PER, I-PER]" + "[AL-AIN, United, Arab, ...]", "[B-LOC, B-LOC, I-LOC, ...]" + "[...]", "[...]" + + """ + + def __init__(self): + headers = [ + 'raw_words', 'target', + ] + super().__init__(headers=headers, indexes=[0, 3]) + + def _load(self, path): + """ + 传入的一个文件路径，将该文件读入DataSet中，field由ConllLoader初始化时指定的headers决定。 + + :param str path: 文件的路径 + :return: DataSet + """ + ds = DataSet() + for idx, data in _read_conll(path, indexes=self.indexes, dropna=self.dropna): + doc_start = False + for i, h in enumerate(self.headers): + field = data[i] + if str(field[0]).startswith('-DOCSTART-'): + doc_start = True + break + if doc_start: + continue + ins = {h: data[i] for i, h in enumerate(self.headers)} + ds.append(Instance(**ins)) + return ds + + def download(self): + raise RuntimeError("conll2003 cannot be downloaded automatically.") + + +class OntoNotesNERLoader(ConllLoader): + """ + 用以读取OntoNotes的NER数据，同时也是Conll2012的NER任务数据。将OntoNote数据处理为conll格式的过程可以参考 + https://github.com/yhcc/OntoNotes-5.0-NER。OntoNoteNERLoader将取第4列和第11列的内容。 + + 返回的DataSet的内容为 + + .. csv-table:: 下面是使用OntoNoteNERLoader读取的DataSet所具备的结构, target列是BIO编码 + :header: "raw_words", "target" + + "[Nadim, Ladki]", "[B-PER, I-PER]" + "[AL-AIN, United, Arab, ...]", "[B-LOC, B-LOC, I-LOC, ...]" + "[...]", "[...]" + + """ + + def __init__(self): + super().__init__(headers=[Const.RAW_WORD, Const.TARGET], indexes=[3, 10]) + + def _load(self, path: str): + dataset = super()._load(path) + + def convert_to_bio(tags): + bio_tags = [] + flag = None + for tag in tags: + label = tag.strip("()*") + if '(' in tag: + bio_label = 'B-' + label + flag = label + elif flag: + bio_label = 'I-' + flag + else: + bio_label = 'O' + if ')' in tag: + flag = None + bio_tags.append(bio_label) + return bio_tags + + def convert_word(words): + converted_words = [] + for word in words: + word = word.replace('/.', '.') # 有些结尾的.是/.形式的 + if not word.startswith('-'): + converted_words.append(word) + continue + # 以下是由于这些符号被转义了，再转回来 + tfrs = {'-LRB-': '(', + '-RRB-': ')', + '-LSB-': '[', + '-RSB-': ']', + '-LCB-': '{', + '-RCB-': '}' + } + if word in tfrs: + converted_words.append(tfrs[word]) + else: + converted_words.append(word) + return converted_words + + dataset.apply_field(convert_word, field_name=Const.RAW_WORD, new_field_name=Const.RAW_WORD) + dataset.apply_field(convert_to_bio, field_name=Const.TARGET, new_field_name=Const.TARGET) + + return dataset + + def download(self): + raise RuntimeError("Ontonotes cannot be downloaded automatically, you can refer " + "https://github.com/yhcc/OntoNotes-5.0-NER to download and preprocess.") + + +class CTBLoader(Loader): + def __init__(self): + super().__init__() + + def _load(self, path: str): + pass + + +class CNNERLoader(Loader): + def _load(self, path: str): + """ + 支持加载形如以下格式的内容，一行两列，以空格隔开两个sample + + Example:: + + 我 O + 们 O + 变 O + 而 O + 以 O + 书 O + 会 O + ... + + :param str path: 文件路径 + :return: DataSet，包含raw_words列和target列 + """ + ds = DataSet() + with open(path, 'r', encoding='utf-8') as f: + raw_chars = [] + target = [] + for line in f: + line = line.strip() + if line: + parts = line.split() + if len(parts) == 1: # 网上下载的数据有一些列少tag，默认补充O + parts.append('O') + raw_chars.append(parts[0]) + target.append(parts[1]) + else: + if raw_chars: + ds.append(Instance(raw_chars=raw_chars, target=target)) + raw_chars = [] + target = [] + return ds + + +class MsraNERLoader(CNNERLoader): + """ + 读取MSRA-NER数据，数据中的格式应该类似与下列的内容 + + Example:: + + 我 O + 们 O + 变 O + 而 O + 以 O + 书 O + 会 O + ... + + 读取后的DataSet包含以下的field + + .. csv-table:: target列是基于BIO的编码方式 + :header: "raw_chars", "target" + + "[我, 们, 变...]", "[O, O, ...]" + "[中, 共, 中, ...]", "[B-ORG, I-ORG, I-ORG, ...]" + "[...]", "[...]" + + """ + + def __init__(self): + super().__init__() + + def download(self, dev_ratio: float = 0.1, re_download: bool = False) -> str: + """ + 自动下载MSAR-NER的数据，如果你使用该数据，请引用 Gina-Anne Levow, 2006, The Third International Chinese Language + Processing Bakeoff: Word Segmentation and Named Entity Recognition. + + 根据dev_ratio的值随机将train中的数据取出一部分作为dev数据。下载完成后在output_dir中有train.conll, test.conll, + dev.conll三个文件。 + + :param float dev_ratio: 如果路径中没有dev集，从train划分多少作为dev的数据. 如果为0，则不划分dev。 + :param bool re_download: 是否重新下载数据，以重新切分数据。 + :return: str, 数据集的目录地址 + :return: + """ + dataset_name = 'msra-ner' + data_dir = self._get_dataset_path(dataset_name=dataset_name) + modify_time = 0 + for filepath in glob.glob(os.path.join(data_dir, '*')): + modify_time = os.stat(filepath).st_mtime + break + if time.time() - modify_time > 1 and re_download: # 通过这种比较丑陋的方式判断一下文件是否是才下载的 + shutil.rmtree(data_dir) + data_dir = self._get_dataset_path(dataset_name=dataset_name) + + if not os.path.exists(os.path.join(data_dir, 'dev.conll')): + if dev_ratio > 0: + assert 0 < dev_ratio < 1, "dev_ratio should be in range (0,1)." + try: + with open(os.path.join(data_dir, 'train.conll'), 'r', encoding='utf-8') as f, \ + open(os.path.join(data_dir, 'middle_file.conll'), 'w', encoding='utf-8') as f1, \ + open(os.path.join(data_dir, 'dev.conll'), 'w', encoding='utf-8') as f2: + lines = [] # 一个sample包含很多行 + for line in f: + line = line.strip() + if line: + lines.append(line) + else: + if random.random() < dev_ratio: + f2.write('\n'.join(lines) + '\n\n') + else: + f1.write('\n'.join(lines) + '\n\n') + lines.clear() + os.remove(os.path.join(data_dir, 'train.conll')) + os.renames(os.path.join(data_dir, 'middle_file.conll'), os.path.join(data_dir, 'train.conll')) + finally: + if os.path.exists(os.path.join(data_dir, 'middle_file.conll')): + os.remove(os.path.join(data_dir, 'middle_file.conll')) + + return data_dir + + +class WeiboNERLoader(CNNERLoader): + def __init__(self): + super().__init__() + + def download(self) -> str: + """ + 自动下载Weibo-NER的数据，如果你使用了该数据，请引用 Nanyun Peng and Mark Dredze, 2015, Named Entity Recognition for + Chinese Social Media with Jointly Trained Embeddings. + + :return: str + """ + dataset_name = 'weibo-ner' + data_dir = self._get_dataset_path(dataset_name=dataset_name) + + return data_dir + + +class PeopleDailyNERLoader(CNNERLoader): + """ + 支持加载的数据格式如下 + + Example:: + + 当 O + 希 O + 望 O + 工 O + 程 O + 救 O + 助 O + 的 O + 百 O + + 读取后的DataSet包含以下的field + + .. csv-table:: target列是基于BIO的编码方式 + :header: "raw_chars", "target" + + "[我, 们, 变...]", "[O, O, ...]" + "[中, 共, 中, ...]", "[B-ORG, I-ORG, I-ORG, ...]" + "[...]", "[...]" + + """ + + def __init__(self): + super().__init__() + + def download(self) -> str: + dataset_name = 'peopledaily' + data_dir = self._get_dataset_path(dataset_name=dataset_name) + + return data_dir diff --git a/fastNLP/io/loader/csv.py b/fastNLP/io/loader/csv.py new file mode 100644 index 00000000..aaf38c00 --- /dev/null +++ b/fastNLP/io/loader/csv.py @@ -0,0 +1,36 @@ +"""undocumented""" + +__all__ = [ + "CSVLoader", +] + +from .loader import Loader +from ..file_reader import _read_csv +from ...core.dataset import DataSet +from ...core.instance import Instance + + +class CSVLoader(Loader): + """ + 读取CSV格式的数据集, 返回 ``DataSet`` 。 + + :param List[str] headers: CSV文件的文件头.定义每一列的属性名称,即返回的DataSet中`field`的名称 + 若为 ``None`` ,则将读入文件的第一行视作 ``headers`` . Default: ``None`` + :param str sep: CSV文件中列与列之间的分隔符. Default: "," + :param bool dropna: 是否忽略非法数据,若 ``True`` 则忽略,若 ``False`` ,在遇到非法数据时,抛出 ``ValueError`` . + Default: ``False`` + """ + + def __init__(self, headers=None, sep=",", dropna=False): + super().__init__() + self.headers = headers + self.sep = sep + self.dropna = dropna + + def _load(self, path): + ds = DataSet() + for idx, data in _read_csv(path, headers=self.headers, + sep=self.sep, dropna=self.dropna): + ds.append(Instance(**data)) + return ds + diff --git a/fastNLP/io/loader/cws.py b/fastNLP/io/loader/cws.py new file mode 100644 index 00000000..2fbb1091 --- /dev/null +++ b/fastNLP/io/loader/cws.py @@ -0,0 +1,94 @@ +"""undocumented""" + +__all__ = [ + "CWSLoader" +] + +import glob +import os +import random +import shutil +import time + +from .loader import Loader +from ...core.dataset import DataSet +from ...core.instance import Instance + + +class CWSLoader(Loader): + """ + CWSLoader支持的数据格式为，一行一句话，不同词之间用空格隔开, 例如： + + Example:: + + 上海 浦东 开发 与 法制 建设 同步 + 新华社 上海 二月 十日 电 （ 记者 谢金虎 、 张持坚 ） + ... + + 该Loader读取后的DataSet具有如下的结构 + + .. csv-table:: + :header: "raw_words" + + "上海 浦东 开发 与 法制 建设 同步" + "新华社 上海 二月 十日 电 （ 记者 谢金虎 、 张持坚 ）" + "..." + + :param: str dataset_name: data的名称，支持pku, msra, cityu(繁体), as(繁体), None + """ + def __init__(self, dataset_name:str=None): + super().__init__() + datanames = {'pku': 'cws-pku', 'msra':'cws-msra', 'as':'cws-as', 'cityu':'cws-cityu'} + if dataset_name in datanames: + self.dataset_name = datanames[dataset_name] + else: + self.dataset_name = None + + def _load(self, path:str): + ds = DataSet() + with open(path, 'r', encoding='utf-8') as f: + for line in f: + line = line.strip() + if line: + ds.append(Instance(raw_words=line)) + return ds + + def download(self, dev_ratio=0.1, re_download=False)->str: + """ + 如果你使用了该数据集，请引用以下的文章:Thomas Emerson, The Second International Chinese Word Segmentation Bakeoff, + 2005. 更多信息可以在http://sighan.cs.uchicago.edu/bakeoff2005/查看 + + :param float dev_ratio: 如果路径中没有dev集，从train划分多少作为dev的数据. 如果为0，则不划分dev。 + :param bool re_download: 是否重新下载数据，以重新切分数据。 + :return: str + """ + if self.dataset_name is None: + return None + data_dir = self._get_dataset_path(dataset_name=self.dataset_name) + modify_time = 0 + for filepath in glob.glob(os.path.join(data_dir, '*')): + modify_time = os.stat(filepath).st_mtime + break + if time.time() - modify_time > 1 and re_download: # 通过这种比较丑陋的方式判断一下文件是否是才下载的 + shutil.rmtree(data_dir) + data_dir = self._get_dataset_path(dataset_name=self.dataset_name) + + if not os.path.exists(os.path.join(data_dir, 'dev.txt')): + if dev_ratio > 0: + assert 0 < dev_ratio < 1, "dev_ratio should be in range (0,1)." + try: + with open(os.path.join(data_dir, 'train.txt'), 'r', encoding='utf-8') as f, \ + open(os.path.join(data_dir, 'middle_file.txt'), 'w', encoding='utf-8') as f1, \ + open(os.path.join(data_dir, 'dev.txt'), 'w', encoding='utf-8') as f2: + for line in f: + if random.random() < dev_ratio: + f2.write(line) + else: + f1.write(line) + os.remove(os.path.join(data_dir, 'train.txt')) + os.renames(os.path.join(data_dir, 'middle_file.txt'), os.path.join(data_dir, 'train.txt')) + finally: + if os.path.exists(os.path.join(data_dir, 'middle_file.txt')): + os.remove(os.path.join(data_dir, 'middle_file.txt')) + + return data_dir diff --git a/fastNLP/io/loader/json.py b/fastNLP/io/loader/json.py new file mode 100644 index 00000000..671769fe --- /dev/null +++ b/fastNLP/io/loader/json.py @@ -0,0 +1,44 @@ +"""undocumented""" + +__all__ = [ + "JsonLoader" +] + +from .loader import Loader +from ..file_reader import _read_json +from ...core.dataset import DataSet +from ...core.instance import Instance + + +class JsonLoader(Loader): + """ + 读取json格式数据.数据必须按行存储,每行是一个包含各类属性的json对象 + + :param dict fields: 需要读入的json属性名称, 和读入后在DataSet中存储的field_name + ``fields`` 的 `key` 必须是json对象的属性名. ``fields`` 的 `value` 为读入后在DataSet存储的 `field_name` , + `value` 也可为 ``None`` , 这时读入后的 `field_name` 与json对象对应属性同名 + ``fields`` 可为 ``None`` , 这时,json对象所有属性都保存在DataSet中. Default: ``None`` + :param bool dropna: 是否忽略非法数据,若 ``True`` 则忽略,若 ``False`` ,在遇到非法数据时,抛出 ``ValueError`` . + Default: ``False`` + """ + + def __init__(self, fields=None, dropna=False): + super(JsonLoader, self).__init__() + self.dropna = dropna + self.fields = None + self.fields_list = None + if fields: + self.fields = {} + for k, v in fields.items(): + self.fields[k] = k if v is None else v + self.fields_list = list(self.fields.keys()) + + def _load(self, path): + ds = DataSet() + for idx, d in _read_json(path, fields=self.fields_list, dropna=self.dropna): + if self.fields: + ins = {self.fields[k]: v for k, v in d.items()} + else: + ins = d + ds.append(Instance(**ins)) + return ds diff --git a/fastNLP/io/loader/loader.py b/fastNLP/io/loader/loader.py new file mode 100644 index 00000000..22636a27 --- /dev/null +++ b/fastNLP/io/loader/loader.py @@ -0,0 +1,91 @@ +"""undocumented""" + +__all__ = [ + "Loader" +] + +from typing import Union, Dict + +from .. import DataBundle +from ..file_utils import _get_dataset_url, get_cache_path, cached_path +from ..utils import check_loader_paths +from ...core.dataset import DataSet + + +class Loader: + """ + 各种数据 Loader 的基类，提供了 API 的参考. + + """ + + def __init__(self): + pass + + def _load(self, path: str) -> DataSet: + """ + 给定一个路径，返回读取的DataSet。 + + :param str path: 路径 + :return: DataSet + """ + raise NotImplementedError + + def load(self, paths: Union[str, Dict[str, str]] = None) -> DataBundle: + """ + 从指定一个或多个路径中的文件中读取数据，返回 :class:`~fastNLP.io.DataBundle` 。 + + 读取的field根据ConllLoader初始化时传入的headers决定。 + + :param Union[str, Dict[str, str]] paths: 支持以下的几种输入方式 + (0) 如果为None，则先查看本地是否有缓存，如果没有则自动下载并缓存。 + + (1) 传入一个目录, 该目录下名称包含train的被认为是train，包含test的被认为是test，包含dev的被认为是dev，如果检测到多个文件 + 名包含'train'、 'dev'、 'test'则会报错:: + + data_bundle = ConllLoader().load('/path/to/dir') # 返回的DataBundle中datasets根据目录下是否检测到train、 + # dev、 test等有所变化，可以通过以下的方式取出DataSet + tr_data = data_bundle.datasets['train'] + te_data = data_bundle.datasets['test'] # 如果目录下有文件包含test这个字段 + + (2) 传入文件路径:: + + data_bundle = ConllLoader().load("/path/to/a/train.conll") # 返回DataBundle对象, datasets中仅包含'train' + tr_data = data_bundle.datasets['train'] # 可以通过以下的方式取出DataSet + + (3) 传入一个dict，比如train，dev，test不在同一个目录下，或者名称中不包含train, dev, test:: + + paths = {'train':"/path/to/tr.conll", 'dev':"/to/validate.conll", "test":"/to/te.conll"} + data_bundle = ConllLoader().load(paths) # 返回的DataBundle中的dataset中包含"train", "dev", "test" + dev_data = data_bundle.datasets['dev'] + + :return: 返回的 :class:`~fastNLP.io.DataBundle` + """ + if paths is None: + paths = self.download() + paths = check_loader_paths(paths) + datasets = {name: self._load(path) for name, path in paths.items()} + data_bundle = DataBundle(datasets=datasets) + return data_bundle + + def download(self) -> str: + """ + 自动下载该数据集 + + :return: 下载后解压目录 + """ + raise NotImplementedError(f"{self.__class__} cannot download data automatically.") + + @staticmethod + def _get_dataset_path(dataset_name): + """ + 传入dataset的名称，获取读取数据的目录。如果数据不存在，会尝试自动下载并缓存 + + :param str dataset_name: 数据集的名称 + :return: str, 数据集的目录地址。直接到该目录下读取相应的数据即可。 + """ + + default_cache_path = get_cache_path() + url = _get_dataset_url(dataset_name) + output_dir = cached_path(url_or_filename=url, cache_dir=default_cache_path, name='dataset') + + return output_dir diff --git a/fastNLP/io/loader/matching.py b/fastNLP/io/loader/matching.py new file mode 100644 index 00000000..a21d0845 --- /dev/null +++ b/fastNLP/io/loader/matching.py @@ -0,0 +1,319 @@ +"""undocumented""" + +__all__ = [ + "MNLILoader", + "SNLILoader", + "QNLILoader", + "RTELoader", + "QuoraLoader", +] + +import os +import warnings +from typing import Union, Dict + +from .json import JsonLoader +from .loader import Loader +from .. import DataBundle +from ...core.const import Const +from ...core.dataset import DataSet +from ...core.instance import Instance + + +class MNLILoader(Loader): + """ + 读取MNLI任务的数据，读取之后的DataSet中包含以下的内容，words0是sentence1, words1是sentence2, target是gold_label, 测试集中没 + 有target列。 + + .. csv-table:: + :header: "raw_words1", "raw_words2", "target" + + "The new rights are...", "Everyone really likes..", "neutral" + "This site includes a...", "The Government Executive...", "contradiction" + "...", "...","." + + """ + + def __init__(self): + super().__init__() + + def _load(self, path: str): + ds = DataSet() + with open(path, 'r', encoding='utf-8') as f: + f.readline() # 跳过header + if path.endswith("test_matched.tsv") or path.endswith('test_mismatched.tsv'): + warnings.warn("RTE's test file has no target.") + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[8] + raw_words2 = parts[9] + if raw_words1 and raw_words2: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2)) + else: + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[8] + raw_words2 = parts[9] + target = parts[-1] + if raw_words1 and raw_words2 and target: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2, target=target)) + return ds + + def load(self, paths: str = None): + """ + + :param str paths: 传入数据所在目录，会在该目录下寻找dev_matched.tsv, dev_mismatched.tsv, test_matched.tsv, + test_mismatched.tsv, train.tsv文件夹 + :return: DataBundle + """ + if paths: + paths = os.path.abspath(os.path.expanduser(paths)) + else: + paths = self.download() + if not os.path.isdir(paths): + raise NotADirectoryError(f"{paths} is not a valid directory.") + + files = {'dev_matched': "dev_matched.tsv", + "dev_mismatched": "dev_mismatched.tsv", + "test_matched": "test_matched.tsv", + "test_mismatched": "test_mismatched.tsv", + "train": 'train.tsv'} + + datasets = {} + for name, filename in files.items(): + filepath = os.path.join(paths, filename) + if not os.path.isfile(filepath): + if 'test' not in name: + raise FileNotFoundError(f"{name} not found in directory {filepath}.") + datasets[name] = self._load(filepath) + + data_bundle = DataBundle(datasets=datasets) + + return data_bundle + + def download(self): + """ + 如果你使用了这个数据，请引用 + + https://www.nyu.edu/projects/bowman/multinli/paper.pdf + :return: + """ + output_dir = self._get_dataset_path('mnli') + return output_dir + + +class SNLILoader(JsonLoader): + """ + 读取之后的DataSet中的field情况为 + + .. csv-table:: 下面是使用SNLILoader加载的DataSet所具备的field + :header: "raw_words1", "raw_words2", "target" + + "The new rights are...", "Everyone really likes..", "neutral" + "This site includes a...", "The Government Executive...", "entailment" + "...", "...", "." + + """ + + def __init__(self): + super().__init__(fields={ + 'sentence1': Const.RAW_WORDS(0), + 'sentence2': Const.RAW_WORDS(1), + 'gold_label': Const.TARGET, + }) + + def load(self, paths: Union[str, Dict[str, str]] = None) -> DataBundle: + """ + 从指定一个或多个路径中的文件中读取数据，返回:class:`~fastNLP.io.DataBundle` 。 + + 读取的field根据ConllLoader初始化时传入的headers决定。 + + :param str paths: 传入一个目录, 将在该目录下寻找snli_1.0_train.jsonl, snli_1.0_dev.jsonl + 和snli_1.0_test.jsonl三个文件。 + + :return: 返回的:class:`~fastNLP.io.DataBundle` + """ + _paths = {} + if paths is None: + paths = self.download() + if paths: + if os.path.isdir(paths): + if not os.path.isfile(os.path.join(paths, 'snli_1.0_train.jsonl')): + raise FileNotFoundError(f"snli_1.0_train.jsonl is not found in {paths}") + _paths['train'] = os.path.join(paths, 'snli_1.0_train.jsonl') + for filename in ['snli_1.0_dev.jsonl', 'snli_1.0_test.jsonl']: + filepath = os.path.join(paths, filename) + _paths[filename.split('_')[-1].split('.')[0]] = filepath + paths = _paths + else: + raise NotADirectoryError(f"{paths} is not a valid directory.") + + datasets = {name: self._load(path) for name, path in paths.items()} + data_bundle = DataBundle(datasets=datasets) + return data_bundle + + def download(self): + """ + 如果您的文章使用了这份数据，请引用 + + http://nlp.stanford.edu/pubs/snli_paper.pdf + + :return: str + """ + return self._get_dataset_path('snli') + + +class QNLILoader(JsonLoader): + """ + QNLI数据集的Loader, + 加载的DataSet将具备以下的field, raw_words1是question, raw_words2是sentence, target是label + + .. csv-table:: + :header: "raw_words1", "raw_words2", "target" + + "What came into force after the new...", "As of that day...", "entailment" + "What is the first major...", "The most important tributaries", "not_entailment" + "...","." + + test数据集没有target列 + + """ + + def __init__(self): + super().__init__() + + def _load(self, path): + ds = DataSet() + + with open(path, 'r', encoding='utf-8') as f: + f.readline() # 跳过header + if path.endswith("test.tsv"): + warnings.warn("QNLI's test file has no target.") + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[1] + raw_words2 = parts[2] + if raw_words1 and raw_words2: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2)) + else: + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[1] + raw_words2 = parts[2] + target = parts[-1] + if raw_words1 and raw_words2 and target: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2, target=target)) + return ds + + def download(self): + """ + 如果您的实验使用到了该数据，请引用 + + .. todo:: + 补充 + + :return: + """ + return self._get_dataset_path('qnli') + + +class RTELoader(Loader): + """ + RTE数据的loader + 加载的DataSet将具备以下的field, raw_words1是sentence0，raw_words2是sentence1, target是label + + .. csv-table:: + :header: "raw_words1", "raw_words2", "target" + + "Dana Reeve, the widow of the actor...", "Christopher Reeve had an...", "not_entailment" + "Yet, we now are discovering that...", "Bacteria is winning...", "entailment" + "...","." + + test数据集没有target列 + """ + + def __init__(self): + super().__init__() + + def _load(self, path: str): + ds = DataSet() + + with open(path, 'r', encoding='utf-8') as f: + f.readline() # 跳过header + if path.endswith("test.tsv"): + warnings.warn("RTE's test file has no target.") + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[1] + raw_words2 = parts[2] + if raw_words1 and raw_words2: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2)) + else: + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[1] + raw_words2 = parts[2] + target = parts[-1] + if raw_words1 and raw_words2 and target: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2, target=target)) + return ds + + def download(self): + return self._get_dataset_path('rte') + + +class QuoraLoader(Loader): + """ + Quora matching任务的数据集Loader + + 支持读取的文件中的内容，应该有以下的形式, 以制表符分隔，且前三列的内容必须是：第一列是label，第二列和第三列是句子 + + Example:: + + 1 How do I get funding for my web based startup idea ? How do I get seed funding pre product ? 327970 + 1 How can I stop my depression ? What can I do to stop being depressed ? 339556 + ... + + 加载的DataSet将具备以下的field + + .. csv-table:: + :header: "raw_words1", "raw_words2", "target" + + "What should I do to avoid...", "1" + "How do I not sleep in a boring class...", "0" + "...","." + + """ + + def __init__(self): + super().__init__() + + def _load(self, path: str): + ds = DataSet() + + with open(path, 'r', encoding='utf-8') as f: + for line in f: + line = line.strip() + if line: + parts = line.split('\t') + raw_words1 = parts[1] + raw_words2 = parts[2] + target = parts[0] + if raw_words1 and raw_words2 and target: + ds.append(Instance(raw_words1=raw_words1, raw_words2=raw_words2, target=target)) + return ds + + def download(self): + raise RuntimeError("Quora cannot be downloaded automatically.") diff --git a/fastNLP/io/model_io.py b/fastNLP/io/model_io.py index ffaa4ef5..9da921df 100644 --- a/fastNLP/io/model_io.py +++ b/fastNLP/io/model_io.py @@ -8,13 +8,9 @@ __all__ = [ import torch -from .base_loader import BaseLoader - -class ModelLoader(BaseLoader): +class ModelLoader: """ - 别名：:class:`fastNLP.io.ModelLoader` :class:`fastNLP.io.model_io.ModelLoader` - 用于读取模型 """ @@ -43,8 +39,6 @@ class ModelLoader(BaseLoader): class ModelSaver(object): """ - 别名：:class:`fastNLP.io.ModelSaver` :class:`fastNLP.io.model_io.ModelSaver` - 用于保存模型 Example:: diff --git a/fastNLP/io/pipe/__init__.py b/fastNLP/io/pipe/__init__.py new file mode 100644 index 00000000..943709e7 --- /dev/null +++ b/fastNLP/io/pipe/__init__.py @@ -0,0 +1,49 @@ +""" +Pipe用于处理通过 Loader 读取的数据，所有的 Pipe 都包含 ``process`` 和 ``process_from_file`` 两种方法。 +``process(data_bundle)`` 传入一个 :class:`~fastNLP.io.DataBundle` 类型的对象, 在传入的 `data_bundle` 上进行原位修改，并将其返回； +``process_from_file(paths)`` 传入的文件路径，返回一个 :class:`~fastNLP.io.DataBundle` 类型的对象。 +``process(data_bundle)`` 或者 ``process_from_file(paths)`` 的返回 `data_bundle` 中的 :class:`~fastNLP.DataSet` +一般都包含原文与转换为index的输入以及转换为index的target；除了 :class:`~fastNLP.DataSet` 之外， +`data_bundle` 还会包含将field转为index时所建立的词表。 + +""" +__all__ = [ + "Pipe", + + "CWSPipe", + + "YelpFullPipe", + "YelpPolarityPipe", + "SSTPipe", + "SST2Pipe", + "IMDBPipe", + "ChnSentiCorpPipe", + + "Conll2003NERPipe", + "OntoNotesNERPipe", + "MsraNERPipe", + "WeiboNERPipe", + "PeopleDailyPipe", + "Conll2003Pipe", + + "MatchingBertPipe", + "RTEBertPipe", + "SNLIBertPipe", + "QuoraBertPipe", + "QNLIBertPipe", + "MNLIBertPipe", + "MatchingPipe", + "RTEPipe", + "SNLIPipe", + "QuoraPipe", + "QNLIPipe", + "MNLIPipe", +] + +from .classification import YelpFullPipe, YelpPolarityPipe, SSTPipe, SST2Pipe, IMDBPipe, ChnSentiCorpPipe +from .conll import Conll2003NERPipe, OntoNotesNERPipe, MsraNERPipe, WeiboNERPipe, PeopleDailyPipe +from .matching import MatchingBertPipe, RTEBertPipe, SNLIBertPipe, QuoraBertPipe, QNLIBertPipe, MNLIBertPipe, \ + MatchingPipe, RTEPipe, SNLIPipe, QuoraPipe, QNLIPipe, MNLIPipe +from .pipe import Pipe +from .conll import Conll2003Pipe +from .cws import CWSPipe diff --git a/fastNLP/io/pipe/classification.py b/fastNLP/io/pipe/classification.py new file mode 100644 index 00000000..3834a570 --- /dev/null +++ b/fastNLP/io/pipe/classification.py @@ -0,0 +1,552 @@ +"""undocumented""" + +__all__ = [ + "YelpFullPipe", + "YelpPolarityPipe", + "SSTPipe", + "SST2Pipe", + 'IMDBPipe', + "ChnSentiCorpPipe" +] + +import re + +from nltk import Tree + +from .pipe import Pipe +from .utils import get_tokenizer, _indexize, _add_words_field, _drop_empty_instance, _add_chars_field +from ..data_bundle import DataBundle +from ..loader.classification import IMDBLoader, YelpFullLoader, SSTLoader, SST2Loader, YelpPolarityLoader +from ...core.const import Const +from ...core.dataset import DataSet +from ...core.instance import Instance +from ...core.vocabulary import Vocabulary +from ..loader.classification import ChnSentiCorpLoader + +nonalpnum = re.compile('[^0-9a-zA-Z?!\']+') + + +class _CLSPipe(Pipe): + """ + 分类问题的基类，负责对classification的数据进行tokenize操作。默认是对raw_words列操作，然后生成words列 + + """ + + def __init__(self, tokenizer: str = 'spacy', lang='en'): + self.tokenizer = get_tokenizer(tokenizer, lang=lang) + + def _tokenize(self, data_bundle, field_name=Const.INPUT, new_field_name=None): + """ + 将DataBundle中的数据进行tokenize + + :param DataBundle data_bundle: + :param str field_name: + :param str new_field_name: + :return: 传入的DataBundle对象 + """ + new_field_name = new_field_name or field_name + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(self.tokenizer, field_name=field_name, new_field_name=new_field_name) + + return data_bundle + + def _granularize(self, data_bundle, tag_map): + """ + 该函数对data_bundle中'target'列中的内容进行转换。 + + :param data_bundle: + :param dict tag_map: 将target列中的tag做以下的映射，比如{"0":0, "1":0, "3":1, "4":1}, 则会删除target为"2"的instance， + 且将"1"认为是第0类。 + :return: 传入的data_bundle + """ + for name in list(data_bundle.datasets.keys()): + dataset = data_bundle.get_dataset(name) + dataset.apply_field(lambda target: tag_map.get(target, -100), field_name=Const.TARGET, + new_field_name=Const.TARGET) + dataset.drop(lambda ins: ins[Const.TARGET] == -100) + data_bundle.set_dataset(dataset, name) + return data_bundle + + +def _clean_str(words): + """ + heavily borrowed from github + https://github.com/LukeZhuang/Hierarchical-Attention-Network/blob/master/yelp-preprocess.ipynb + :param sentence: is a str + :return: + """ + words_collection = [] + for word in words: + if word in ['-lrb-', '-rrb-', '', '-r', '-l', 'b-']: + continue + tt = nonalpnum.split(word) + t = ''.join(tt) + if t != '': + words_collection.append(t) + + return words_collection + + +class YelpFullPipe(_CLSPipe): + """ + 处理YelpFull的数据, 处理之后DataSet中的内容如下 + + .. csv-table:: 下面是使用YelpFullPipe处理后的DataSet所具备的field + :header: "raw_words", "words", "target", "seq_len" + + "It 's a ...", "[4, 2, 10, ...]", 0, 10 + "Offers that ...", "[20, 40, ...]", 1, 21 + "...", "[...]", ., . + + :param bool lower: 是否对输入进行小写化。 + :param int granularity: 支持2, 3, 5。若为2, 则认为是2分类问题，将1、2归为1类，4、5归为一类，丢掉2；若为3, 则有3分类问题，将 + 1、2归为1类，3归为1类，4、5归为1类；若为5, 则有5分类问题。 + :param str tokenizer: 使用哪种tokenize方式将数据切成单词。支持'spacy'和'raw'。raw使用空格作为切分。 + """ + + def __init__(self, lower: bool = False, granularity=5, tokenizer: str = 'spacy'): + super().__init__(tokenizer=tokenizer, lang='en') + self.lower = lower + assert granularity in (2, 3, 5), "granularity can only be 2,3,5." + self.granularity = granularity + + if granularity == 2: + self.tag_map = {"1": 0, "2": 0, "4": 1, "5": 1} + elif granularity == 3: + self.tag_map = {"1": 0, "2": 0, "3": 1, "4": 2, "5": 2} + else: + self.tag_map = {"1": 0, "2": 1, "3": 2, "4": 3, "5": 4} + + def _tokenize(self, data_bundle, field_name=Const.INPUT, new_field_name=None): + """ + 将DataBundle中的数据进行tokenize + + :param DataBundle data_bundle: + :param str field_name: + :param str new_field_name: + :return: 传入的DataBundle对象 + """ + new_field_name = new_field_name or field_name + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(self.tokenizer, field_name=field_name, new_field_name=new_field_name) + dataset.apply_field(_clean_str, field_name=field_name, new_field_name=new_field_name) + return data_bundle + + def process(self, data_bundle): + """ + 传入的DataSet应该具备如下的结构 + + .. csv-table:: + :header: "raw_words", "target" + + "I got 'new' tires from them and... ", "1" + "Don't waste your time. We had two...", "1" + "...", "..." + + :param data_bundle: + :return: + """ + + # 复制一列words + data_bundle = _add_words_field(data_bundle, lower=self.lower) + + # 进行tokenize + data_bundle = self._tokenize(data_bundle=data_bundle, field_name=Const.INPUT) + + # 根据granularity设置tag + data_bundle = self._granularize(data_bundle, tag_map=self.tag_map) + + # 删除空行 + data_bundle = _drop_empty_instance(data_bundle, field_name=Const.INPUT) + + # index + data_bundle = _indexize(data_bundle=data_bundle) + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + + data_bundle.set_input(Const.INPUT, Const.INPUT_LEN) + data_bundle.set_target(Const.TARGET) + + return data_bundle + + def process_from_file(self, paths=None): + """ + + :param paths: + :return: DataBundle + """ + data_bundle = YelpFullLoader().load(paths) + return self.process(data_bundle=data_bundle) + + +class YelpPolarityPipe(_CLSPipe): + """ + 处理YelpPolarity的数据, 处理之后DataSet中的内容如下 + + .. csv-table:: 下面是使用YelpFullPipe处理后的DataSet所具备的field + :header: "raw_words", "words", "target", "seq_len" + + "It 's a ...", "[4, 2, 10, ...]", 0, 10 + "Offers that ...", "[20, 40, ...]", 1, 21 + "...", "[...]", ., . + + :param bool lower: 是否对输入进行小写化。 + :param str tokenizer: 使用哪种tokenize方式将数据切成单词。支持'spacy'和'raw'。raw使用空格作为切分。 + """ + + def __init__(self, lower: bool = False, tokenizer: str = 'spacy'): + super().__init__(tokenizer=tokenizer, lang='en') + self.lower = lower + + def process(self, data_bundle): + # 复制一列words + data_bundle = _add_words_field(data_bundle, lower=self.lower) + + # 进行tokenize + data_bundle = self._tokenize(data_bundle=data_bundle, field_name=Const.INPUT) + # index + data_bundle = _indexize(data_bundle=data_bundle) + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + + data_bundle.set_input(Const.INPUT, Const.INPUT_LEN) + data_bundle.set_target(Const.TARGET) + + return data_bundle + + def process_from_file(self, paths=None): + """ + + :param str paths: + :return: DataBundle + """ + data_bundle = YelpPolarityLoader().load(paths) + return self.process(data_bundle=data_bundle) + + +class SSTPipe(_CLSPipe): + """ + 经过该Pipe之后，DataSet中具备的field如下所示 + + .. csv-table:: 下面是使用SSTPipe处理后的DataSet所具备的field + :header: "raw_words", "words", "target", "seq_len" + + "It 's a ...", "[4, 2, 10, ...]", 0, 16 + "Offers that ...", "[20, 40, ...]", 1, 18 + "...", "[...]", ., . + + :param bool subtree: 是否将train, test, dev数据展开为子树，扩充数据量。 Default: ``False`` + :param bool train_subtree: 是否将train集通过子树扩展数据。 + :param bool lower: 是否对输入进行小写化。 + :param int granularity: 支持2, 3, 5。若为2, 则认为是2分类问题，将0、1归为1类，3、4归为一类，丢掉2；若为3, 则有3分类问题，将 + 0、1归为1类，2归为1类，3、4归为1类；若为5, 则有5分类问题。 + :param str tokenizer: 使用哪种tokenize方式将数据切成单词。支持'spacy'和'raw'。raw使用空格作为切分。 + """ + + def __init__(self, subtree=False, train_subtree=True, lower=False, granularity=5, tokenizer='spacy'): + super().__init__(tokenizer=tokenizer, lang='en') + self.subtree = subtree + self.train_tree = train_subtree + self.lower = lower + assert granularity in (2, 3, 5), "granularity can only be 2,3,5." + self.granularity = granularity + + if granularity == 2: + self.tag_map = {"0": 0, "1": 0, "3": 1, "4": 1} + elif granularity == 3: + self.tag_map = {"0": 0, "1": 0, "2": 1, "3": 2, "4": 2} + else: + self.tag_map = {"0": 0, "1": 1, "2": 2, "3": 3, "4": 4} + + def process(self, data_bundle: DataBundle): + """ + 对DataBundle中的数据进行预处理。输入的DataSet应该至少拥有raw_words这一列，且内容类似与 + + .. csv-table:: + :header: "raw_words" + + "(3 (2 It) (4 (4 (2 's) (4 (3 (2 a)..." + "(4 (4 (2 Offers) (3 (3 (2 that) (3 (3 rare)..." + "..." + + :param ~fastNLP.io.DataBundle data_bundle: 需要处理的DataBundle对象 + :return: + """ + # 先取出subtree + for name in list(data_bundle.datasets.keys()): + dataset = data_bundle.get_dataset(name) + ds = DataSet() + use_subtree = self.subtree or (name == 'train' and self.train_tree) + for ins in dataset: + raw_words = ins['raw_words'] + tree = Tree.fromstring(raw_words) + if use_subtree: + for t in tree.subtrees(): + raw_words = " ".join(t.leaves()) + instance = Instance(raw_words=raw_words, target=t.label()) + ds.append(instance) + else: + instance = Instance(raw_words=' '.join(tree.leaves()), target=tree.label()) + ds.append(instance) + data_bundle.set_dataset(ds, name) + + _add_words_field(data_bundle, lower=self.lower) + + # 进行tokenize + data_bundle = self._tokenize(data_bundle=data_bundle, field_name=Const.INPUT) + + # 根据granularity设置tag + data_bundle = self._granularize(data_bundle, tag_map=self.tag_map) + + # index + data_bundle = _indexize(data_bundle=data_bundle) + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + + data_bundle.set_input(Const.INPUT, Const.INPUT_LEN) + data_bundle.set_target(Const.TARGET) + + return data_bundle + + def process_from_file(self, paths=None): + data_bundle = SSTLoader().load(paths) + return self.process(data_bundle=data_bundle) + + +class SST2Pipe(_CLSPipe): + """ + 加载SST2的数据, 处理完成之后DataSet将拥有以下的field + + .. csv-table:: + :header: "raw_words", "words", "target", "seq_len" + + "it 's a charming and... ", "[3, 4, 5, 6, 7,...]", 1, 43 + "unflinchingly bleak and...", "[10, 11, 7,...]", 1, 21 + "...", "...", ., . + + :param bool lower: 是否对输入进行小写化。 + :param str tokenizer: 使用哪种tokenize方式将数据切成单词。支持'spacy'和'raw'。raw使用空格作为切分。 + """ + + def __init__(self, lower=False, tokenizer='spacy'): + super().__init__(tokenizer=tokenizer, lang='en') + self.lower = lower + + def process(self, data_bundle: DataBundle): + """ + 可以处理的DataSet应该具备如下的结构 + + .. csv-table:: + :header: "raw_words", "target" + + "it 's a charming and... ", 1 + "unflinchingly bleak and...", 1 + "...", "..." + + :param data_bundle: + :return: + """ + _add_words_field(data_bundle, self.lower) + + data_bundle = self._tokenize(data_bundle=data_bundle) + + src_vocab = Vocabulary() + src_vocab.from_dataset(data_bundle.datasets['train'], field_name=Const.INPUT, + no_create_entry_dataset=[dataset for name, dataset in data_bundle.datasets.items() if + name != 'train']) + src_vocab.index_dataset(*data_bundle.datasets.values(), field_name=Const.INPUT) + + tgt_vocab = Vocabulary(unknown=None, padding=None) + tgt_vocab.from_dataset(data_bundle.datasets['train'], field_name=Const.TARGET) + datasets = [] + for name, dataset in data_bundle.datasets.items(): + if dataset.has_field(Const.TARGET): + datasets.append(dataset) + tgt_vocab.index_dataset(*datasets, field_name=Const.TARGET) + + data_bundle.set_vocab(src_vocab, Const.INPUT) + data_bundle.set_vocab(tgt_vocab, Const.TARGET) + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + + data_bundle.set_input(Const.INPUT, Const.INPUT_LEN) + data_bundle.set_target(Const.TARGET) + + return data_bundle + + def process_from_file(self, paths=None): + """ + + :param str paths: 如果为None，则自动下载并缓存到fastNLP的缓存地址。 + :return: DataBundle + """ + data_bundle = SST2Loader().load(paths) + return self.process(data_bundle) + + +class IMDBPipe(_CLSPipe): + """ + 经过本Pipe处理后DataSet将如下 + + .. csv-table:: 输出DataSet的field + :header: "raw_words", "words", "target", "seq_len" + + "Bromwell High is a cartoon ... ", "[3, 5, 6, 9, ...]", 0, 20 + "Story of a man who has ...", "[20, 43, 9, 10, ...]", 1, 31 + "...", "[...]", ., . + + 其中raw_words为str类型，是原文; words是转换为index的输入; target是转换为index的目标值; + words列被设置为input; target列被设置为target。 + + :param bool lower: 是否将words列的数据小写。 + :param str tokenizer: 使用什么tokenizer来将句子切分为words. 支持spacy, raw两种。raw即使用空格拆分。 + """ + + def __init__(self, lower: bool = False, tokenizer: str = 'spacy'): + super().__init__(tokenizer=tokenizer, lang='en') + self.lower = lower + + def process(self, data_bundle: DataBundle): + """ + 期待的DataBunlde中输入的DataSet应该类似于如下，有两个field，raw_words和target，且均为str类型 + + .. csv-table:: 输入DataSet的field + :header: "raw_words", "target" + + "Bromwell High is a cartoon ... ", "pos" + "Story of a man who has ...", "neg" + "...", "..." + + :param DataBunlde data_bundle: 传入的DataBundle中的DataSet必须包含raw_words和target两个field，且raw_words列应该为str, + target列应该为str。 + :return: DataBundle + """ + + # 替换
+ def replace_br(raw_words): + raw_words = raw_words.replace("
", ' ') + return raw_words + + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(replace_br, field_name=Const.RAW_WORD, new_field_name=Const.RAW_WORD) + + _add_words_field(data_bundle, lower=self.lower) + self._tokenize(data_bundle, field_name=Const.INPUT, new_field_name=Const.INPUT) + _indexize(data_bundle) + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + dataset.set_input(Const.INPUT, Const.INPUT_LEN) + dataset.set_target(Const.TARGET) + + return data_bundle + + def process_from_file(self, paths=None): + """ + + :param paths: 支持路径类型参见 :class:`fastNLP.io.loader.Loader` 的load函数。 + :return: DataBundle + """ + # 读取数据 + data_bundle = IMDBLoader().load(paths) + data_bundle = self.process(data_bundle) + + return data_bundle + + +class ChnSentiCorpPipe(Pipe): + """ + 处理之后的DataSet有以下的结构 + + .. csv-table:: + :header: "raw_chars", "chars", "target", "seq_len" + + "這間酒店環境和服務態度亦算不錯,但房間空間太小~~", "[2, 3, 4, 5, ...]", 1, 31 + "<荐书> 推荐所有喜欢<红楼>...", "[10, 21, ....]", 1, 25 + "..." + + 其中chars, seq_len是input，target是target + + :param bool bigrams: 是否增加一列bigrams. bigrams的构成是['复', '旦', '大', '学', ...]->["复旦", "旦大", ...]。如果 + 设置为True，返回的DataSet将有一列名为bigrams, 且已经转换为了index并设置为input，对应的vocab可以通过 + data_bundle.get_vocab('bigrams')获取. + :param bool trigrams: 是否增加一列trigrams. trigrams的构成是 ['复', '旦', '大', '学', ...]->["复旦大", "旦大学", ...] + 。如果设置为True，返回的DataSet将有一列名为trigrams, 且已经转换为了index并设置为input，对应的vocab可以通过 + data_bundle.get_vocab('trigrams')获取. + """ + def __init__(self, bigrams=False, trigrams=False): + super().__init__() + + self.bigrams = bigrams + self.trigrams = trigrams + + def _tokenize(self, data_bundle): + """ + 将DataSet中的"复旦大学"拆分为["复", "旦", "大", "学"]. 未来可以通过扩展这个函数实现分词。 + + :param data_bundle: + :return: + """ + data_bundle.apply_field(list, field_name=Const.CHAR_INPUT, new_field_name=Const.CHAR_INPUT) + return data_bundle + + def process(self, data_bundle:DataBundle): + """ + 可以处理的DataSet应该具备以下的field + + .. csv-table:: + :header: "raw_chars", "target" + + "這間酒店環境和服務態度亦算不錯,但房間空間太小~~", "1" + "<荐书> 推荐所有喜欢<红楼>...", "1" + "..." + + :param data_bundle: + :return: + """ + _add_chars_field(data_bundle, lower=False) + + data_bundle = self._tokenize(data_bundle) + + input_field_names = [Const.CHAR_INPUT] + if self.bigrams: + for name, dataset in data_bundle.iter_datasets(): + dataset.apply_field(lambda chars: [c1 + c2 for c1, c2 in zip(chars, chars[1:] + [''])], + field_name=Const.CHAR_INPUT, new_field_name='bigrams') + input_field_names.append('bigrams') + if self.trigrams: + for name, dataset in data_bundle.iter_datasets(): + dataset.apply_field(lambda chars: [c1 + c2 + c3 for c1, c2, c3 in + zip(chars, chars[1:] + [''], chars[2:] + [''] * 2)], + field_name=Const.CHAR_INPUT, new_field_name='trigrams') + input_field_names.append('trigrams') + + # index + _indexize(data_bundle, input_field_names, Const.TARGET) + + input_fields = [Const.TARGET, Const.INPUT_LEN] + input_field_names + target_fields = [Const.TARGET] + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.CHAR_INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + + def process_from_file(self, paths=None): + """ + + :param paths: 支持路径类型参见 :class:`fastNLP.io.loader.Loader` 的load函数。 + :return: DataBundle + """ + # 读取数据 + data_bundle = ChnSentiCorpLoader().load(paths) + data_bundle = self.process(data_bundle) + + return data_bundle \ No newline at end of file diff --git a/fastNLP/io/pipe/conll.py b/fastNLP/io/pipe/conll.py new file mode 100644 index 00000000..a96b259a --- /dev/null +++ b/fastNLP/io/pipe/conll.py @@ -0,0 +1,354 @@ +"""undocumented""" + +__all__ = [ + "Conll2003NERPipe", + "Conll2003Pipe", + "OntoNotesNERPipe", + "MsraNERPipe", + "PeopleDailyPipe", + "WeiboNERPipe" +] + +from .pipe import Pipe +from .utils import _add_chars_field +from .utils import _indexize, _add_words_field +from .utils import iob2, iob2bioes +from .. import DataBundle +from ..loader.conll import Conll2003NERLoader, OntoNotesNERLoader +from ..loader.conll import PeopleDailyNERLoader, WeiboNERLoader, MsraNERLoader, ConllLoader +from ...core.const import Const +from ...core.vocabulary import Vocabulary + + +class _NERPipe(Pipe): + """ + NER任务的处理Pipe, 该Pipe会（1）复制raw_words列，并命名为words; (2）在words, target列建立词表 + (创建 :class:`fastNLP.Vocabulary` 对象，所以在返回的DataBundle中将有两个Vocabulary); (3）将words，target列根据相应的 + Vocabulary转换为index。 + + raw_words列为List[str], 是未转换的原始数据; words列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有words, target, seq_len; 设置为target有target, seq_len。 + + :param: str encoding_type: target列使用什么类型的encoding方式，支持bioes, bio两种。 + :param bool lower: 是否将words小写化后再建立词表，绝大多数情况都不需要设置为True。 + """ + + def __init__(self, encoding_type: str = 'bio', lower: bool = False): + if encoding_type == 'bio': + self.convert_tag = iob2 + else: + self.convert_tag = lambda words: iob2bioes(iob2(words)) + self.lower = lower + + def process(self, data_bundle: DataBundle) -> DataBundle: + """ + 支持的DataSet的field为 + + .. csv-table:: + :header: "raw_words", "target" + + "[Nadim, Ladki]", "[B-PER, I-PER]" + "[AL-AIN, United, Arab, ...]", "[B-LOC, B-LOC, I-LOC, ...]" + "[...]", "[...]" + + :param ~fastNLP.DataBundle data_bundle: 传入的DataBundle中的DataSet必须包含raw_words和ner两个field，且两个field的内容均为List[str]。 + 在传入DataBundle基础上原位修改。 + :return: DataBundle + """ + # 转换tag + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(self.convert_tag, field_name=Const.TARGET, new_field_name=Const.TARGET) + + _add_words_field(data_bundle, lower=self.lower) + + # index + _indexize(data_bundle) + + input_fields = [Const.TARGET, Const.INPUT, Const.INPUT_LEN] + target_fields = [Const.TARGET, Const.INPUT_LEN] + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + + +class Conll2003NERPipe(_NERPipe): + """ + Conll2003的NER任务的处理Pipe, 该Pipe会（1）复制raw_words列，并命名为words; (2）在words, target列建立词表 + (创建 :class:`fastNLP.Vocabulary` 对象，所以在返回的DataBundle中将有两个Vocabulary); (3）将words，target列根据相应的 + Vocabulary转换为index。 + 经过该Pipe过后，DataSet中的内容如下所示 + + .. csv-table:: Following is a demo layout of DataSet returned by Conll2003Loader + :header: "raw_words", "words", "target", "seq_len" + + "[Nadim, Ladki]", "[2, 3]", "[1, 2]", 2 + "[AL-AIN, United, Arab, ...]", "[4, 5, 6,...]", "[3, 4,...]", 6 + "[...]", "[...]", "[...]", . + + raw_words列为List[str], 是未转换的原始数据; words列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有words, target, seq_len; 设置为target有target。 + + :param: str encoding_type: target列使用什么类型的encoding方式，支持bioes, bio两种。 + :param bool lower: 是否将words小写化后再建立词表，绝大多数情况都不需要设置为True。 + """ + + def process_from_file(self, paths) -> DataBundle: + """ + + :param paths: 支持路径类型参见 :class:`fastNLP.io.loader.ConllLoader` 的load函数。 + :return: DataBundle + """ + # 读取数据 + data_bundle = Conll2003NERLoader().load(paths) + data_bundle = self.process(data_bundle) + + return data_bundle + + +class Conll2003Pipe(Pipe): + def __init__(self, chunk_encoding_type='bioes', ner_encoding_type='bioes', lower: bool = False): + """ + 经过该Pipe后，DataSet中的内容如下 + + .. csv-table:: + :header: "raw_words", "words", "pos", "chunk", "ner", "seq_len" + + "[Nadim, Ladki]", "[2, 3]", "[0, 0]", "[1, 2]", "[1, 2]", 2 + "[AL-AIN, United, Arab, ...]", "[4, 5, 6,...]", "[1, 2...]", "[3, 4...]", "[3, 4...]", 6 + "[...]", "[...]", "[...]", "[...]", "[...]". + + 其中words, seq_len是input; pos, chunk, ner, seq_len是target + + :param str chunk_encoding_type: 支持bioes, bio。 + :param str ner_encoding_type: 支持bioes, bio。 + :param bool lower: 是否将words列小写化后再建立词表 + """ + if chunk_encoding_type == 'bio': + self.chunk_convert_tag = iob2 + else: + self.chunk_convert_tag = lambda tags: iob2bioes(iob2(tags)) + if ner_encoding_type == 'bio': + self.ner_convert_tag = iob2 + else: + self.ner_convert_tag = lambda tags: iob2bioes(iob2(tags)) + self.lower = lower + + def process(self, data_bundle) -> DataBundle: + """ + 输入的DataSet应该类似于如下的形式 + + .. csv-table:: + :header: "raw_words", "pos", "chunk", "ner" + + "[Nadim, Ladki]", "[NNP, NNP]", "[B-NP, I-NP]", "[B-PER, I-PER]" + "[AL-AIN, United, Arab, ...]", "[NNP, NNP...]", "[B-NP, B-NP, ...]", "[B-LOC, B-LOC,...]" + "[...]", "[...]", "[...]", "[...]". + + :param data_bundle: + :return: 传入的DataBundle + """ + # 转换tag + for name, dataset in data_bundle.datasets.items(): + dataset.drop(lambda x: "-DOCSTART-" in x[Const.RAW_WORD]) + dataset.apply_field(self.chunk_convert_tag, field_name='chunk', new_field_name='chunk') + dataset.apply_field(self.ner_convert_tag, field_name='ner', new_field_name='ner') + + _add_words_field(data_bundle, lower=self.lower) + + # index + _indexize(data_bundle, input_field_names=Const.INPUT, target_field_names=['pos', 'ner']) + # chunk中存在一些tag只在dev中出现，没在train中 + tgt_vocab = Vocabulary(unknown=None, padding=None) + tgt_vocab.from_dataset(*data_bundle.datasets.values(), field_name='chunk') + tgt_vocab.index_dataset(*data_bundle.datasets.values(), field_name='chunk') + data_bundle.set_vocab(tgt_vocab, 'chunk') + + input_fields = [Const.INPUT, Const.INPUT_LEN] + target_fields = ['pos', 'ner', 'chunk', Const.INPUT_LEN] + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + + def process_from_file(self, paths): + """ + + :param paths: + :return: + """ + data_bundle = ConllLoader(headers=['raw_words', 'pos', 'chunk', 'ner']).load(paths) + return self.process(data_bundle) + + +class OntoNotesNERPipe(_NERPipe): + """ + 处理OntoNotes的NER数据，处理之后DataSet中的field情况为 + + .. csv-table:: + :header: "raw_words", "words", "target", "seq_len" + + "[Nadim, Ladki]", "[2, 3]", "[1, 2]", 2 + "[AL-AIN, United, Arab, ...]", "[4, 5, 6,...]", "[3, 4]", 6 + "[...]", "[...]", "[...]", . + + raw_words列为List[str], 是未转换的原始数据; words列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有words, target, seq_len; 设置为target有target。 + + :param: str encoding_type: target列使用什么类型的encoding方式，支持bioes, bio两种。 + :param bool lower: 是否将words小写化后再建立词表，绝大多数情况都不需要设置为True。 + """ + + def process_from_file(self, paths): + data_bundle = OntoNotesNERLoader().load(paths) + return self.process(data_bundle) + + +class _CNNERPipe(Pipe): + """ + 中文NER任务的处理Pipe, 该Pipe会（1）复制raw_chars列，并命名为chars; (2）在chars, target列建立词表 + (创建 :class:`fastNLP.Vocabulary` 对象，所以在返回的DataBundle中将有两个Vocabulary); (3）将chars，target列根据相应的 + Vocabulary转换为index。 + + raw_chars列为List[str], 是未转换的原始数据; chars列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有chars, target, seq_len; 设置为target有target, seq_len。 + + :param: str encoding_type: target列使用什么类型的encoding方式，支持bioes, bio两种。 + :param bool bigrams: 是否增加一列bigrams. bigrams的构成是['复', '旦', '大', '学', ...]->["复旦", "旦大", ...]。如果 + 设置为True，返回的DataSet将有一列名为bigrams, 且已经转换为了index并设置为input，对应的vocab可以通过 + data_bundle.get_vocab('bigrams')获取. + :param bool trigrams: 是否增加一列trigrams. trigrams的构成是 ['复', '旦', '大', '学', ...]->["复旦大", "旦大学", ...] + 。如果设置为True，返回的DataSet将有一列名为trigrams, 且已经转换为了index并设置为input，对应的vocab可以通过 + data_bundle.get_vocab('trigrams')获取. + """ + + def __init__(self, encoding_type: str = 'bio', bigrams=False, trigrams=False): + if encoding_type == 'bio': + self.convert_tag = iob2 + else: + self.convert_tag = lambda words: iob2bioes(iob2(words)) + + self.bigrams = bigrams + self.trigrams = trigrams + + def process(self, data_bundle: DataBundle) -> DataBundle: + """ + 支持的DataSet的field为 + + .. csv-table:: + :header: "raw_chars", "target" + + "[相, 比, 之, 下,...]", "[O, O, O, O, ...]" + "[青, 岛, 海, 牛, 队, 和, ...]", "[B-ORG, I-ORG, I-ORG, ...]" + "[...]", "[...]" + + raw_chars列为List[str], 是未转换的原始数据; chars列为List[int]，是转换为index的输入数据; target列是List[int]， + 是转换为index的target。返回的DataSet中被设置为input有chars, target, seq_len; 设置为target有target。 + + :param ~fastNLP.DataBundle data_bundle: 传入的DataBundle中的DataSet必须包含raw_words和ner两个field，且两个field + 的内容均为List[str]。在传入DataBundle基础上原位修改。 + :return: DataBundle + """ + # 转换tag + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(self.convert_tag, field_name=Const.TARGET, new_field_name=Const.TARGET) + + _add_chars_field(data_bundle, lower=False) + + input_field_names = [Const.CHAR_INPUT] + if self.bigrams: + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars: [c1 + c2 for c1, c2 in zip(chars, chars[1:] + [''])], + field_name=Const.CHAR_INPUT, new_field_name='bigrams') + input_field_names.append('bigrams') + if self.trigrams: + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars: [c1 + c2 + c3 for c1, c2, c3 in + zip(chars, chars[1:] + [''], chars[2:] + [''] * 2)], + field_name=Const.CHAR_INPUT, new_field_name='trigrams') + input_field_names.append('trigrams') + + # index + _indexize(data_bundle, input_field_names, Const.TARGET) + + input_fields = [Const.TARGET, Const.INPUT_LEN] + input_field_names + target_fields = [Const.TARGET, Const.INPUT_LEN] + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.CHAR_INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + + +class MsraNERPipe(_CNNERPipe): + """ + 处理MSRA-NER的数据，处理之后的DataSet的field情况为 + + .. csv-table:: + :header: "raw_chars", "chars", "target", "seq_len" + + "[相, 比, 之, 下,...]", "[2, 3, 4, 5, ...]", "[0, 0, 0, 0, ...]", 11 + "[青, 岛, 海, 牛, 队, 和, ...]", "[10, 21, ....]", "[1, 2, 3, ...]", 21 + "[...]", "[...]", "[...]", . + + raw_chars列为List[str], 是未转换的原始数据; chars列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有chars, target, seq_len; 设置为target有target。 + + """ + + def process_from_file(self, paths=None) -> DataBundle: + data_bundle = MsraNERLoader().load(paths) + return self.process(data_bundle) + + +class PeopleDailyPipe(_CNNERPipe): + """ + 处理people daily的ner的数据，处理之后的DataSet的field情况为 + + .. csv-table:: + :header: "raw_chars", "chars", "target", "seq_len" + + "[相, 比, 之, 下,...]", "[2, 3, 4, 5, ...]", "[0, 0, 0, 0, ...]", 11 + "[青, 岛, 海, 牛, 队, 和, ...]", "[10, 21, ....]", "[1, 2, 3, ...]", 21 + "[...]", "[...]", "[...]", . + + raw_chars列为List[str], 是未转换的原始数据; chars列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有chars, target, seq_len; 设置为target有target。 + """ + + def process_from_file(self, paths=None) -> DataBundle: + data_bundle = PeopleDailyNERLoader().load(paths) + return self.process(data_bundle) + + +class WeiboNERPipe(_CNNERPipe): + """ + 处理weibo的ner的数据，处理之后的DataSet的field情况为 + + .. csv-table:: + :header: "raw_chars", "chars", "target", "seq_len" + + "[相, 比, 之, 下,...]", "[2, 3, 4, 5, ...]", "[0, 0, 0, 0, ...]", 11 + "[青, 岛, 海, 牛, 队, 和, ...]", "[10, 21, ....]", "[1, 2, 3, ...]", 21 + "[...]", "[...]", "[...]", . + + raw_chars列为List[str], 是未转换的原始数据; chars列为List[int]，是转换为index的输入数据; target列是List[int]，是转换为index的 + target。返回的DataSet中被设置为input有chars, target, seq_len; 设置为target有target。 + + :param: str encoding_type: target列使用什么类型的encoding方式，支持bioes, bio两种。 + """ + + def process_from_file(self, paths=None) -> DataBundle: + data_bundle = WeiboNERLoader().load(paths) + return self.process(data_bundle) diff --git a/fastNLP/io/pipe/cws.py b/fastNLP/io/pipe/cws.py new file mode 100644 index 00000000..748cf10a --- /dev/null +++ b/fastNLP/io/pipe/cws.py @@ -0,0 +1,266 @@ +"""undocumented""" + +__all__ = [ + "CWSPipe" +] + +import re +from itertools import chain + +from .pipe import Pipe +from .utils import _indexize +from .. import DataBundle +from ..loader import CWSLoader +from ...core.const import Const + + +def _word_lens_to_bmes(word_lens): + """ + + :param list word_lens: List[int], 每个词语的长度 + :return: List[str], BMES的序列 + """ + tags = [] + for word_len in word_lens: + if word_len == 1: + tags.append('S') + else: + tags.append('B') + tags.extend(['M'] * (word_len - 2)) + tags.append('E') + return tags + + +def _word_lens_to_segapp(word_lens): + """ + + :param list word_lens: List[int], 每个词语的长度 + :return: List[str], BMES的序列 + """ + tags = [] + for word_len in word_lens: + if word_len == 1: + tags.append('SEG') + else: + tags.extend(['APP'] * (word_len - 1)) + tags.append('SEG') + return tags + + +def _alpha_span_to_special_tag(span): + """ + 将span替换成特殊的字符 + + :param str span: + :return: + """ + if 'oo' == span.lower(): # speical case when represent 2OO8 + return span + if len(span) == 1: + return span + else: + return '' + + +def _find_and_replace_alpha_spans(line): + """ + 传入原始句子，替换其中的字母为特殊标记 + + :param str line:原始数据 + :return: str + """ + new_line = '' + pattern = '[a-zA-Z]+(?=[\u4e00-\u9fff ，％,.。！<－“])' + prev_end = 0 + for match in re.finditer(pattern, line): + start, end = match.span() + span = line[start:end] + new_line += line[prev_end:start] + _alpha_span_to_special_tag(span) + prev_end = end + new_line += line[prev_end:] + return new_line + + +def _digit_span_to_special_tag(span): + """ + + :param str span: 需要替换的str + :return: + """ + if span[0] == '0' and len(span) > 2: + return '' + decimal_point_count = 0 # one might have more than one decimal pointers + for idx, char in enumerate(span): + if char == '.' or char == '﹒' or char == '·': + decimal_point_count += 1 + if span[-1] == '.' or span[-1] == '﹒' or span[ + -1] == '·': # last digit being decimal point means this is not a number + if decimal_point_count == 1: + return span + else: + return '' + if decimal_point_count == 1: + return '' + elif decimal_point_count > 1: + return '' + else: + return '' + + +def _find_and_replace_digit_spans(line): + """ + only consider words start with number, contains '.', characters. + + If ends with space, will be processed + + If ends with Chinese character, will be processed + + If ends with or contains english char, not handled. + + floats are replaced by + + otherwise unkdgt + """ + new_line = '' + pattern = '\d[\d\\.﹒·]*(?=[\u4e00-\u9fff ，％,。！<－“])' + prev_end = 0 + for match in re.finditer(pattern, line): + start, end = match.span() + span = line[start:end] + new_line += line[prev_end:start] + _digit_span_to_special_tag(span) + prev_end = end + new_line += line[prev_end:] + return new_line + + +class CWSPipe(Pipe): + """ + 对CWS数据进行预处理, 处理之后的数据，具备以下的结构 + + .. csv-table:: + :header: "raw_words", "chars", "target", "bigrams", "trigrams", "seq_len" + + "共同 创造 美好...", "[2, 3, 4...]", "[0, 2, 0, 2,...]", "[10, 4, 1,...]","[6, 4, 1,...]", 13 + "2001年 新年 钟声...", "[8, 9, 9, 7, ...]", "[0, 1, 1, 1, 2...]", "[11, 12, ...]","[3, 9, ...]", 20 + "...", "[...]","[...]", "[...]","[...]", . + + 其中bigrams仅当bigrams列为True的时候为真 + + :param str,None dataset_name: 支持'pku', 'msra', 'cityu', 'as', None + :param str encoding_type: 可以选择'bmes', 'segapp'两种。"我 来自 复旦大学...", bmes的tag为[S, B, E, B, M, M, E...]; segapp + 的tag为[seg, app, seg, app, app, app, seg, ...] + :param bool replace_num_alpha: 是否将数字和字母用特殊字符替换。 + :param bool bigrams: 是否增加一列bigram. bigram的构成是['复', '旦', '大', '学', ...]->["复旦", "旦大", ...] + :param bool trigrams: 是否增加一列trigram. trigram的构成是 ['复', '旦', '大', '学', ...]->["复旦大", "旦大学", ...] + """ + + def __init__(self, dataset_name=None, encoding_type='bmes', replace_num_alpha=True, bigrams=False, trigrams=False): + if encoding_type == 'bmes': + self.word_lens_to_tags = _word_lens_to_bmes + else: + self.word_lens_to_tags = _word_lens_to_segapp + + self.dataset_name = dataset_name + self.bigrams = bigrams + self.trigrams = trigrams + self.replace_num_alpha = replace_num_alpha + + def _tokenize(self, data_bundle): + """ + 将data_bundle中的'chars'列切分成一个一个的word. + 例如输入是"共同 创造 美好.."->[[共, 同], [创, 造], [...], ] + + :param data_bundle: + :return: + """ + def split_word_into_chars(raw_chars): + words = raw_chars.split() + chars = [] + for word in words: + char = [] + subchar = [] + for c in word: + if c == '<': + subchar.append(c) + continue + if c == '>' and subchar[0] == '<': + char.append(''.join(subchar)) + subchar = [] + if subchar: + subchar.append(c) + else: + char.append(c) + char.extend(subchar) + chars.append(char) + return chars + + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(split_word_into_chars, field_name=Const.CHAR_INPUT, + new_field_name=Const.CHAR_INPUT) + return data_bundle + + def process(self, data_bundle: DataBundle) -> DataBundle: + """ + 可以处理的DataSet需要包含raw_words列 + + .. csv-table:: + :header: "raw_words" + + "上海 浦东 开发 与 法制 建设 同步" + "新华社 上海 二月 十日 电 （ 记者 谢金虎 、 张持坚 ）" + "..." + + :param data_bundle: + :return: + """ + data_bundle.copy_field(Const.RAW_WORD, Const.CHAR_INPUT) + + if self.replace_num_alpha: + data_bundle.apply_field(_find_and_replace_alpha_spans, Const.CHAR_INPUT, Const.CHAR_INPUT) + data_bundle.apply_field(_find_and_replace_digit_spans, Const.CHAR_INPUT, Const.CHAR_INPUT) + + self._tokenize(data_bundle) + + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars: self.word_lens_to_tags(map(len, chars)), field_name=Const.CHAR_INPUT, + new_field_name=Const.TARGET) + dataset.apply_field(lambda chars: list(chain(*chars)), field_name=Const.CHAR_INPUT, + new_field_name=Const.CHAR_INPUT) + input_field_names = [Const.CHAR_INPUT] + if self.bigrams: + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars: [c1 + c2 for c1, c2 in zip(chars, chars[1:] + [''])], + field_name=Const.CHAR_INPUT, new_field_name='bigrams') + input_field_names.append('bigrams') + if self.trigrams: + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars: [c1 + c2 + c3 for c1, c2, c3 in + zip(chars, chars[1:] + [''], chars[2:] + [''] * 2)], + field_name=Const.CHAR_INPUT, new_field_name='trigrams') + input_field_names.append('trigrams') + + _indexize(data_bundle, input_field_names, Const.TARGET) + + input_fields = [Const.TARGET, Const.INPUT_LEN] + input_field_names + target_fields = [Const.TARGET, Const.INPUT_LEN] + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.CHAR_INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + + def process_from_file(self, paths=None) -> DataBundle: + """ + + :param str paths: + :return: + """ + if self.dataset_name is None and paths is None: + raise RuntimeError( + "You have to set `paths` when calling process_from_file() or `dataset_name `when initialization.") + if self.dataset_name is not None and paths is not None: + raise RuntimeError("You cannot specify `paths` and `dataset_name` simultaneously") + data_bundle = CWSLoader(self.dataset_name).load(paths) + return self.process(data_bundle) diff --git a/fastNLP/io/pipe/matching.py b/fastNLP/io/pipe/matching.py new file mode 100644 index 00000000..747e7b44 --- /dev/null +++ b/fastNLP/io/pipe/matching.py @@ -0,0 +1,274 @@ +"""undocumented""" + +__all__ = [ + "MatchingBertPipe", + "RTEBertPipe", + "SNLIBertPipe", + "QuoraBertPipe", + "QNLIBertPipe", + "MNLIBertPipe", + "MatchingPipe", + "RTEPipe", + "SNLIPipe", + "QuoraPipe", + "QNLIPipe", + "MNLIPipe", +] + +from .pipe import Pipe +from .utils import get_tokenizer +from ..loader.matching import SNLILoader, MNLILoader, QNLILoader, RTELoader, QuoraLoader +from ...core.const import Const +from ...core.vocabulary import Vocabulary + + +class MatchingBertPipe(Pipe): + """ + Matching任务的Bert pipe，输出的DataSet将包含以下的field + + .. csv-table:: + :header: "raw_words1", "raw_words2", "words", "target", "seq_len" + + "The new rights are...", "Everyone really likes..", "[2, 3, 4, 5, ...]", 1, 10 + "This site includes a...", "The Government Executive...", "[11, 12, 13,...]", 0, 5 + "...", "...", "[...]", ., . + + words列是将raw_words1(即premise), raw_words2(即hypothesis)使用"[SEP]"链接起来转换为index的。 + words列被设置为input，target列被设置为target和input(设置为input以方便在forward函数中计算loss， + 如果不在forward函数中计算loss也不影响，fastNLP将根据forward函数的形参名进行传参). + + :param bool lower: 是否将word小写化。 + :param str tokenizer: 使用什么tokenizer来将句子切分为words. 支持spacy, raw两种。raw即使用空格拆分。 + """ + + def __init__(self, lower=False, tokenizer: str = 'raw'): + super().__init__() + + self.lower = bool(lower) + self.tokenizer = get_tokenizer(tokenizer=tokenizer) + + def _tokenize(self, data_bundle, field_names, new_field_names): + """ + + :param DataBundle data_bundle: DataBundle. + :param list field_names: List[str], 需要tokenize的field名称 + :param list new_field_names: List[str], tokenize之后field的名称，与field_names一一对应。 + :return: 输入的DataBundle对象 + """ + for name, dataset in data_bundle.datasets.items(): + for field_name, new_field_name in zip(field_names, new_field_names): + dataset.apply_field(lambda words: self.tokenizer(words), field_name=field_name, + new_field_name=new_field_name) + return data_bundle + + def process(self, data_bundle): + for dataset in data_bundle.datasets.values(): + if dataset.has_field(Const.TARGET): + dataset.drop(lambda x: x[Const.TARGET] == '-') + + for name, dataset in data_bundle.datasets.items(): + dataset.copy_field(Const.RAW_WORDS(0), Const.INPUTS(0), ) + dataset.copy_field(Const.RAW_WORDS(1), Const.INPUTS(1), ) + + if self.lower: + for name, dataset in data_bundle.datasets.items(): + dataset[Const.INPUTS(0)].lower() + dataset[Const.INPUTS(1)].lower() + + data_bundle = self._tokenize(data_bundle, [Const.INPUTS(0), Const.INPUTS(1)], + [Const.INPUTS(0), Const.INPUTS(1)]) + + # concat两个words + def concat(ins): + words0 = ins[Const.INPUTS(0)] + words1 = ins[Const.INPUTS(1)] + words = words0 + ['[SEP]'] + words1 + return words + + for name, dataset in data_bundle.datasets.items(): + dataset.apply(concat, new_field_name=Const.INPUT) + dataset.delete_field(Const.INPUTS(0)) + dataset.delete_field(Const.INPUTS(1)) + + word_vocab = Vocabulary() + word_vocab.from_dataset(*[dataset for name, dataset in data_bundle.datasets.items() if 'train' in name], + field_name=Const.INPUT, + no_create_entry_dataset=[dataset for name, dataset in data_bundle.datasets.items() if + 'train' not in name]) + word_vocab.index_dataset(*data_bundle.datasets.values(), field_name=Const.INPUT) + + target_vocab = Vocabulary(padding=None, unknown=None) + target_vocab.from_dataset(data_bundle.datasets['train'], field_name=Const.TARGET) + has_target_datasets = [dataset for name, dataset in data_bundle.datasets.items() if + dataset.has_field(Const.TARGET)] + target_vocab.index_dataset(*has_target_datasets, field_name=Const.TARGET) + + data_bundle.set_vocab(word_vocab, Const.INPUT) + data_bundle.set_vocab(target_vocab, Const.TARGET) + + input_fields = [Const.INPUT, Const.INPUT_LEN] + target_fields = [Const.TARGET] + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUT) + dataset.set_input(*input_fields, flag=True) + for fields in target_fields: + if dataset.has_field(fields): + dataset.set_target(fields, flag=True) + + return data_bundle + + +class RTEBertPipe(MatchingBertPipe): + def process_from_file(self, paths=None): + data_bundle = RTELoader().load(paths) + return self.process(data_bundle) + + +class SNLIBertPipe(MatchingBertPipe): + def process_from_file(self, paths=None): + data_bundle = SNLILoader().load(paths) + return self.process(data_bundle) + + +class QuoraBertPipe(MatchingBertPipe): + def process_from_file(self, paths): + data_bundle = QuoraLoader().load(paths) + return self.process(data_bundle) + + +class QNLIBertPipe(MatchingBertPipe): + def process_from_file(self, paths=None): + data_bundle = QNLILoader().load(paths) + return self.process(data_bundle) + + +class MNLIBertPipe(MatchingBertPipe): + def process_from_file(self, paths=None): + data_bundle = MNLILoader().load(paths) + return self.process(data_bundle) + + +class MatchingPipe(Pipe): + """ + Matching任务的Pipe。输出的DataSet将包含以下的field + + .. csv-table:: + :header: "raw_words1", "raw_words2", "words1", "words2", "target", "seq_len1", "seq_len2" + + "The new rights are...", "Everyone really likes..", "[2, 3, 4, 5, ...]", "[10, 20, 6]", 1, 10, 13 + "This site includes a...", "The Government Executive...", "[11, 12, 13,...]", "[2, 7, ...]", 0, 6, 7 + "...", "...", "[...]", "[...]", ., ., . + + words1是premise，words2是hypothesis。其中words1,words2,seq_len1,seq_len2被设置为input；target被设置为target + 和input(设置为input以方便在forward函数中计算loss，如果不在forward函数中计算loss也不影响，fastNLP将根据forward函数 + 的形参名进行传参)。 + + :param bool lower: 是否将所有raw_words转为小写。 + :param str tokenizer: 将原始数据tokenize的方式。支持spacy, raw. spacy是使用spacy切分，raw就是用空格切分。 + """ + + def __init__(self, lower=False, tokenizer: str = 'raw'): + super().__init__() + + self.lower = bool(lower) + self.tokenizer = get_tokenizer(tokenizer=tokenizer) + + def _tokenize(self, data_bundle, field_names, new_field_names): + """ + + :param ~fastNLP.DataBundle data_bundle: DataBundle. + :param list field_names: List[str], 需要tokenize的field名称 + :param list new_field_names: List[str], tokenize之后field的名称，与field_names一一对应。 + :return: 输入的DataBundle对象 + """ + for name, dataset in data_bundle.datasets.items(): + for field_name, new_field_name in zip(field_names, new_field_names): + dataset.apply_field(lambda words: self.tokenizer(words), field_name=field_name, + new_field_name=new_field_name) + return data_bundle + + def process(self, data_bundle): + """ + 接受的DataBundle中的DataSet应该具有以下的field, target列可以没有 + + .. csv-table:: + :header: "raw_words1", "raw_words2", "target" + + "The new rights are...", "Everyone really likes..", "entailment" + "This site includes a...", "The Government Executive...", "not_entailment" + "...", "..." + + :param ~fastNLP.DataBundle data_bundle: 通过loader读取得到的data_bundle，里面包含了数据集的原始数据内容 + :return: data_bundle + """ + data_bundle = self._tokenize(data_bundle, [Const.RAW_WORDS(0), Const.RAW_WORDS(1)], + [Const.INPUTS(0), Const.INPUTS(1)]) + + for dataset in data_bundle.datasets.values(): + if dataset.has_field(Const.TARGET): + dataset.drop(lambda x: x[Const.TARGET] == '-') + + if self.lower: + for name, dataset in data_bundle.datasets.items(): + dataset[Const.INPUTS(0)].lower() + dataset[Const.INPUTS(1)].lower() + + word_vocab = Vocabulary() + word_vocab.from_dataset(*[dataset for name, dataset in data_bundle.datasets.items() if 'train' in name], + field_name=[Const.INPUTS(0), Const.INPUTS(1)], + no_create_entry_dataset=[dataset for name, dataset in data_bundle.datasets.items() if + 'train' not in name]) + word_vocab.index_dataset(*data_bundle.datasets.values(), field_name=[Const.INPUTS(0), Const.INPUTS(1)]) + + target_vocab = Vocabulary(padding=None, unknown=None) + target_vocab.from_dataset(data_bundle.datasets['train'], field_name=Const.TARGET) + has_target_datasets = [dataset for name, dataset in data_bundle.datasets.items() if + dataset.has_field(Const.TARGET)] + target_vocab.index_dataset(*has_target_datasets, field_name=Const.TARGET) + + data_bundle.set_vocab(word_vocab, Const.INPUTS(0)) + data_bundle.set_vocab(target_vocab, Const.TARGET) + + input_fields = [Const.INPUTS(0), Const.INPUTS(1), Const.INPUT_LENS(0), Const.INPUT_LENS(1)] + target_fields = [Const.TARGET] + + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.INPUTS(0), Const.INPUT_LENS(0)) + dataset.add_seq_len(Const.INPUTS(1), Const.INPUT_LENS(1)) + dataset.set_input(*input_fields, flag=True) + for fields in target_fields: + if dataset.has_field(fields): + dataset.set_target(fields, flag=True) + + return data_bundle + + +class RTEPipe(MatchingPipe): + def process_from_file(self, paths=None): + data_bundle = RTELoader().load(paths) + return self.process(data_bundle) + + +class SNLIPipe(MatchingPipe): + def process_from_file(self, paths=None): + data_bundle = SNLILoader().load(paths) + return self.process(data_bundle) + + +class QuoraPipe(MatchingPipe): + def process_from_file(self, paths): + data_bundle = QuoraLoader().load(paths) + return self.process(data_bundle) + + +class QNLIPipe(MatchingPipe): + def process_from_file(self, paths=None): + data_bundle = QNLILoader().load(paths) + return self.process(data_bundle) + + +class MNLIPipe(MatchingPipe): + def process_from_file(self, paths=None): + data_bundle = MNLILoader().load(paths) + return self.process(data_bundle) diff --git a/fastNLP/io/pipe/pipe.py b/fastNLP/io/pipe/pipe.py new file mode 100644 index 00000000..db65ece6 --- /dev/null +++ b/fastNLP/io/pipe/pipe.py @@ -0,0 +1,32 @@ +"""undocumented""" + +__all__ = [ + "Pipe", +] + +from .. import DataBundle + + +class Pipe: + """ + .. todo:: + doc + + """ + def process(self, data_bundle: DataBundle) -> DataBundle: + """ + 对输入的DataBundle进行处理，然后返回该DataBundle。 + + :param ~fastNLP.DataBundle data_bundle: 需要处理的DataBundle对象 + :return: + """ + raise NotImplementedError + + def process_from_file(self, paths) -> DataBundle: + """ + 传入文件路径，生成处理好的DataBundle对象。paths支持的路径形式可以参考 `fastNLP.io.loader.Loader.load()` + + :param paths: + :return: DataBundle + """ + raise NotImplementedError diff --git a/fastNLP/io/pipe/utils.py b/fastNLP/io/pipe/utils.py new file mode 100644 index 00000000..ea7e0aa8 --- /dev/null +++ b/fastNLP/io/pipe/utils.py @@ -0,0 +1,176 @@ +"""undocumented""" + +__all__ = [ + "iob2", + "iob2bioes", + "get_tokenizer", +] + +from typing import List + +from ...core.const import Const +from ...core.vocabulary import Vocabulary + + +def iob2(tags: List[str]) -> List[str]: + """ + 检查数据是否是合法的IOB数据，如果是IOB1会被自动转换为IOB2。两种格式的区别见 + https://datascience.stackexchange.com/questions/37824/difference-between-iob-and-iob2-format + + :param tags: 需要转换的tags + """ + for i, tag in enumerate(tags): + if tag == "O": + continue + split = tag.split("-") + if len(split) != 2 or split[0] not in ["I", "B"]: + raise TypeError("The encoding schema is not a valid IOB type.") + if split[0] == "B": + continue + elif i == 0 or tags[i - 1] == "O": # conversion IOB1 to IOB2 + tags[i] = "B" + tag[1:] + elif tags[i - 1][1:] == tag[1:]: + continue + else: # conversion IOB1 to IOB2 + tags[i] = "B" + tag[1:] + return tags + + +def iob2bioes(tags: List[str]) -> List[str]: + """ + 将iob的tag转换为bioes编码 + :param tags: + :return: + """ + new_tags = [] + for i, tag in enumerate(tags): + if tag == 'O': + new_tags.append(tag) + else: + split = tag.split('-')[0] + if split == 'B': + if i + 1 != len(tags) and tags[i + 1].split('-')[0] == 'I': + new_tags.append(tag) + else: + new_tags.append(tag.replace('B-', 'S-')) + elif split == 'I': + if i + 1 < len(tags) and tags[i + 1].split('-')[0] == 'I': + new_tags.append(tag) + else: + new_tags.append(tag.replace('I-', 'E-')) + else: + raise TypeError("Invalid IOB format.") + return new_tags + + +def get_tokenizer(tokenizer: str, lang='en'): + """ + + :param str tokenizer: 获取tokenzier方法 + :param str lang: 语言，当前仅支持en + :return: 返回tokenize函数 + """ + if tokenizer == 'spacy': + import spacy + spacy.prefer_gpu() + if lang != 'en': + raise RuntimeError("Spacy only supports en right right.") + en = spacy.load(lang) + tokenizer = lambda x: [w.text for w in en.tokenizer(x)] + elif tokenizer == 'raw': + tokenizer = _raw_split + else: + raise RuntimeError("Only support `spacy`, `raw` tokenizer.") + return tokenizer + + +def _raw_split(sent): + return sent.split() + + +def _indexize(data_bundle, input_field_names=Const.INPUT, target_field_names=Const.TARGET): + """ + 在dataset中的field_name列建立词表，Const.TARGET列建立词表，并把词表加入到data_bundle中。 + + :param ~fastNLP.DataBundle data_bundle: + :param: str,list input_field_names: + :param: str,list target_field_names: 这一列的vocabulary没有unknown和padding + :return: + """ + if isinstance(input_field_names, str): + input_field_names = [input_field_names] + if isinstance(target_field_names, str): + target_field_names = [target_field_names] + for input_field_name in input_field_names: + src_vocab = Vocabulary() + src_vocab.from_dataset(data_bundle.datasets['train'], field_name=input_field_name, + no_create_entry_dataset=[dataset for name, dataset in data_bundle.datasets.items() if + name != 'train']) + src_vocab.index_dataset(*data_bundle.datasets.values(), field_name=input_field_name) + data_bundle.set_vocab(src_vocab, input_field_name) + + for target_field_name in target_field_names: + tgt_vocab = Vocabulary(unknown=None, padding=None) + tgt_vocab.from_dataset(data_bundle.datasets['train'], field_name=target_field_name) + tgt_vocab.index_dataset(*data_bundle.datasets.values(), field_name=target_field_name) + data_bundle.set_vocab(tgt_vocab, target_field_name) + + return data_bundle + + +def _add_words_field(data_bundle, lower=False): + """ + 给data_bundle中的dataset中复制一列words. 并根据lower参数判断是否需要小写化 + + :param data_bundle: + :param bool lower:是否要小写化 + :return: 传入的DataBundle + """ + data_bundle.copy_field(field_name=Const.RAW_WORD, new_field_name=Const.INPUT, ignore_miss_dataset=True) + + if lower: + for name, dataset in data_bundle.datasets.items(): + dataset[Const.INPUT].lower() + return data_bundle + + +def _add_chars_field(data_bundle, lower=False): + """ + 给data_bundle中的dataset中复制一列chars. 并根据lower参数判断是否需要小写化 + + :param data_bundle: + :param bool lower:是否要小写化 + :return: 传入的DataBundle + """ + data_bundle.copy_field(field_name=Const.RAW_CHAR, new_field_name=Const.CHAR_INPUT, ignore_miss_dataset=True) + + if lower: + for name, dataset in data_bundle.datasets.items(): + dataset[Const.CHAR_INPUT].lower() + return data_bundle + + +def _drop_empty_instance(data_bundle, field_name): + """ + 删除data_bundle的DataSet中存在的某个field为空的情况 + + :param ~fastNLP.DataBundle data_bundle: + :param str field_name: 对哪个field进行检查，如果为None，则任意field为空都会删掉 + :return: 传入的DataBundle + """ + + def empty_instance(ins): + if field_name: + field_value = ins[field_name] + if field_value in ((), {}, [], ''): + return True + return False + for _, field_value in ins.items(): + if field_value in ((), {}, [], ''): + return True + return False + + for name, dataset in data_bundle.datasets.items(): + dataset.drop(empty_instance) + + return data_bundle diff --git a/fastNLP/io/utils.py b/fastNLP/io/utils.py index a7d2de85..e1de2ae7 100644 --- a/fastNLP/io/utils.py +++ b/fastNLP/io/utils.py @@ -1,23 +1,37 @@ -import os +""" +.. todo:: + doc +""" + +__all__ = [ + "check_loader_paths" +] +import os +from pathlib import Path from typing import Union, Dict +from ..core import logger + -def check_dataloader_paths(paths:Union[str, Dict[str, str]])->Dict[str, str]: +def check_loader_paths(paths: Union[str, Dict[str, str]]) -> Dict[str, str]: """ - 检查传入dataloader的文件的合法性。如果为合法路径，将返回至少包含'train'这个key的dict。类似于下面的结果 - { - 'train': '/some/path/to/', # 一定包含，建词表应该在这上面建立，剩下的其它文件应该只需要处理并index。 - 'test': 'xxx' # 可能有，也可能没有 - ... - } - 如果paths为不合法的，将直接进行raise相应的错误 + 检查传入dataloader的文件的合法性。如果为合法路径，将返回至少包含'train'这个key的dict。类似于下面的结果:: - :param paths: 路径. 可以为一个文件路径(则认为该文件就是train的文件); 可以为一个文件目录，将在该目录下寻找train(文件名 + { + 'train': '/some/path/to/', # 一定包含，建词表应该在这上面建立，剩下的其它文件应该只需要处理并index。 + 'test': 'xxx' # 可能有，也可能没有 + ... + } + + 如果paths为不合法的，将直接进行raise相应的错误. 如果paths内不包含train也会报错。 + + :param str paths: 路径. 可以为一个文件路径(则认为该文件就是train的文件); 可以为一个文件目录，将在该目录下寻找train(文件名 中包含train这个字段), test.txt, dev.txt; 可以为一个dict, 则key是用户自定义的某个文件的名称，value是这个文件的路径。 :return: """ - if isinstance(paths, str): + if isinstance(paths, (str, Path)): + paths = os.path.abspath(os.path.expanduser(paths)) if os.path.isfile(paths): return {'train': paths} elif os.path.isdir(paths): @@ -29,26 +43,32 @@ def check_dataloader_paths(paths:Union[str, Dict[str, str]])->Dict[str, str]: path_pair = ('train', filename) if 'dev' in filename: if path_pair: - raise Exception("File:{} in {} contains bot `{}` and `dev`.".format(filename, paths, path_pair[0])) + raise Exception( + "File:{} in {} contains bot `{}` and `dev`.".format(filename, paths, path_pair[0])) path_pair = ('dev', filename) if 'test' in filename: if path_pair: - raise Exception("File:{} in {} contains bot `{}` and `test`.".format(filename, paths, path_pair[0])) + raise Exception( + "File:{} in {} contains bot `{}` and `test`.".format(filename, paths, path_pair[0])) path_pair = ('test', filename) if path_pair: files[path_pair[0]] = os.path.join(paths, path_pair[1]) + if 'train' not in files: + raise KeyError(f"There is no train file in {paths}.") return files else: raise FileNotFoundError(f"{paths} is not a valid file path.") - + elif isinstance(paths, dict): if paths: if 'train' not in paths: raise KeyError("You have to include `train` in your dict.") for key, value in paths.items(): if isinstance(key, str) and isinstance(value, str): + value = os.path.abspath(os.path.expanduser(value)) if not os.path.isfile(value): raise TypeError(f"{value} is not a valid file.") + paths[key] = value else: raise TypeError("All keys and values in paths should be str.") return paths @@ -57,13 +77,14 @@ def check_dataloader_paths(paths:Union[str, Dict[str, str]])->Dict[str, str]: else: raise TypeError(f"paths only supports str and dict. not {type(paths)}.") + def get_tokenizer(): try: import spacy spacy.prefer_gpu() en = spacy.load('en') - print('use spacy tokenizer') + logger.info('use spacy tokenizer') return lambda x: [w.text for w in en.tokenizer(x)] except Exception as e: - print('use raw tokenizer') + logger.error('use raw tokenizer') return lambda x: x.split() diff --git a/fastNLP/models/__init__.py b/fastNLP/models/__init__.py index 14314049..62adbf69 100644 --- a/fastNLP/models/__init__.py +++ b/fastNLP/models/__init__.py @@ -21,14 +21,24 @@ __all__ = [ "STSeqCls", "BiaffineParser", - "GraphParser" + "GraphParser", + + "BertForSequenceClassification", + "BertForSentenceMatching", + "BertForMultipleChoice", + "BertForTokenClassification", + "BertForQuestionAnswering" ] from .base_model import BaseModel from .bert import BertForMultipleChoice, BertForQuestionAnswering, BertForSequenceClassification, \ - BertForTokenClassification + BertForTokenClassification, BertForSentenceMatching from .biaffine_parser import BiaffineParser, GraphParser from .cnn_text_classification import CNNText from .sequence_labeling import SeqLabeling, AdvSeqLabel from .snli import ESIM from .star_transformer import StarTransEnc, STSeqCls, STNLICls, STSeqLabel + +import sys +from ..doc_utils import doc_process +doc_process(sys.modules[__name__]) \ No newline at end of file diff --git a/fastNLP/models/base_model.py b/fastNLP/models/base_model.py index 2646d580..61edb91f 100644 --- a/fastNLP/models/base_model.py +++ b/fastNLP/models/base_model.py @@ -1,3 +1,7 @@ +"""undocumented""" + +__all__ = [] + import torch from ..modules.decoder.mlp import MLP diff --git a/fastNLP/models/bert.py b/fastNLP/models/bert.py index adecab60..30ed0cd8 100644 --- a/fastNLP/models/bert.py +++ b/fastNLP/models/bert.py @@ -1,338 +1,257 @@ """ -bert.py is modified from huggingface/pytorch-pretrained-BERT, which is licensed under the Apache License 2.0. +fastNLP提供了BERT应用到五个下游任务的模型代码，可以直接调用。这五个任务分别为 + + - 文本分类任务： :class:`~fastNLP.models.BertForSequenceClassification` + - Matching任务： :class:`~fastNLP.models.BertForSentenceMatching` + - 多选任务： :class:`~fastNLP.models.BertForMultipleChoice` + - 序列标注任务： :class:`~fastNLP.models.BertForTokenClassification` + - 抽取式QA任务： :class:`~fastNLP.models.BertForQuestionAnswering` + +每一个模型必须要传入一个名字为 `embed` 的 :class:`fastNLP.embeddings.BertEmbedding` ，这个参数包含了 +:class:`fastNLP.modules.encoder.BertModel` ，是下游模型的编码器(encoder)。 + +除此以外，还需要传入一个数字，这个数字在不同下游任务模型上的意义如下:: + + 下游任务模型 参数名称 含义 + BertForSequenceClassification num_labels 文本分类类别数目，默认值为2 + BertForSentenceMatching num_labels Matching任务类别数目，默认值为2 + BertForMultipleChoice num_choices 多选任务选项数目，默认值为2 + BertForTokenClassification num_labels 序列标注标签数目，无默认值 + BertForQuestionAnswering num_labels 抽取式QA列数，默认值为2(即第一列为start_span, 第二列为end_span) + +最后还可以传入dropout的大小，默认值为0.1。 """ + +__all__ = [ + "BertForSequenceClassification", + "BertForSentenceMatching", + "BertForMultipleChoice", + "BertForTokenClassification", + "BertForQuestionAnswering" +] + +import warnings + import torch from torch import nn from .base_model import BaseModel from ..core.const import Const -from ..modules.encoder import BertModel -from ..modules.encoder.bert import BertConfig +from ..core._logger import logger +from ..embeddings import BertEmbedding class BertForSequenceClassification(BaseModel): - """BERT model for classification. - This module is composed of the BERT model with a linear layer on top of - the pooled output. - Params: - `config`: a BertConfig class instance with the configuration to build a new model. - `num_labels`: the number of classes for the classifier. Default = 2. - Inputs: - `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length] - with the word token indices in the vocabulary. Items in the batch should begin with the special "CLS" token. (see the tokens preprocessing logic in the scripts - `extract_features.py`, `run_classifier.py` and `run_squad.py`) - `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token - types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to - a `sentence B` token (see BERT paper for more details). - `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices - selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max - input sequence length in the current batch. It's the mask that we typically use for attention when - a batch has varying length sentences. - `labels`: labels for the classification output: torch.LongTensor of shape [batch_size] - with indices selected in [0, ..., num_labels]. - Outputs: - if `labels` is not `None`: - Outputs the CrossEntropy classification loss of the output with the labels. - if `labels` is `None`: - Outputs the classification logits of shape [batch_size, num_labels]. - Example usage: - ```python - # Already been converted into WordPiece token ids - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) - config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768, - num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072) - num_labels = 2 - model = BertForSequenceClassification(num_labels, config) - logits = model(input_ids, token_type_ids, input_mask) - ``` """ - def __init__(self, num_labels, config=None, bert_dir=None): + BERT model for classification. + + :param fastNLP.embeddings.BertEmbedding embed: 下游模型的编码器(encoder). + :param int num_labels: 文本分类类别数目，默认值为2. + :param float dropout: dropout的大小，默认值为0.1. + """ + def __init__(self, embed: BertEmbedding, num_labels: int=2, dropout=0.1): super(BertForSequenceClassification, self).__init__() + self.num_labels = num_labels - if bert_dir is not None: - self.bert = BertModel.from_pretrained(bert_dir) - else: - if config is None: - config = BertConfig(30522) - self.bert = BertModel(config) - self.dropout = nn.Dropout(config.hidden_dropout_prob) - self.classifier = nn.Linear(config.hidden_size, num_labels) - - @classmethod - def from_pretrained(cls, num_labels, pretrained_model_dir): - config = BertConfig(pretrained_model_dir) - model = cls(num_labels=num_labels, config=config, bert_dir=pretrained_model_dir) - return model - - def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None): - _, pooled_output = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False) - pooled_output = self.dropout(pooled_output) - logits = self.classifier(pooled_output) + self.bert = embed + self.dropout = nn.Dropout(p=dropout) + self.classifier = nn.Linear(self.bert.embedding_dim, num_labels) - if labels is not None: - loss_fct = nn.CrossEntropyLoss() - loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1)) - return {Const.OUTPUT: logits, Const.LOSS: loss} - else: - return {Const.OUTPUT: logits} + if not self.bert.model.include_cls_sep: + self.bert.model.include_cls_sep = True + warn_msg = "Bert for sequence classification excepts BertEmbedding `include_cls_sep` True, " \ + "but got False. FastNLP has changed it to True." + logger.warn(warn_msg) + warnings.warn(warn_msg) - def predict(self, input_ids, token_type_ids=None, attention_mask=None): - logits = self.forward(input_ids, token_type_ids, attention_mask) + def forward(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.Tensor [batch_size, num_labels] + """ + hidden = self.dropout(self.bert(words)) + cls_hidden = hidden[:, 0] + logits = self.classifier(cls_hidden) + + return {Const.OUTPUT: logits} + + def predict(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.LongTensor [batch_size] + """ + logits = self.forward(words)[Const.OUTPUT] + return {Const.OUTPUT: torch.argmax(logits, dim=-1)} + + +class BertForSentenceMatching(BaseModel): + """ + BERT model for sentence matching. + + :param fastNLP.embeddings.BertEmbedding embed: 下游模型的编码器(encoder). + :param int num_labels: Matching任务类别数目，默认值为2. + :param float dropout: dropout的大小，默认值为0.1. + """ + def __init__(self, embed: BertEmbedding, num_labels: int=2, dropout=0.1): + super(BertForSentenceMatching, self).__init__() + self.num_labels = num_labels + self.bert = embed + self.dropout = nn.Dropout(p=dropout) + self.classifier = nn.Linear(self.bert.embedding_dim, num_labels) + + if not self.bert.model.include_cls_sep: + self.bert.model.include_cls_sep = True + warn_msg = "Bert for sentence matching excepts BertEmbedding `include_cls_sep` True, " \ + "but got False. FastNLP has changed it to True." + logger.warn(warn_msg) + warnings.warn(warn_msg) + + def forward(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.Tensor [batch_size, num_labels] + """ + hidden = self.bert(words) + cls_hidden = self.dropout(hidden[:, 0]) + logits = self.classifier(cls_hidden) + + return {Const.OUTPUT: logits} + + def predict(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.LongTensor [batch_size] + """ + logits = self.forward(words)[Const.OUTPUT] return {Const.OUTPUT: torch.argmax(logits, dim=-1)} class BertForMultipleChoice(BaseModel): - """BERT model for multiple choice tasks. - This module is composed of the BERT model with a linear layer on top of - the pooled output. - Params: - `config`: a BertConfig class instance with the configuration to build a new model. - `num_choices`: the number of classes for the classifier. Default = 2. - Inputs: - `input_ids`: a torch.LongTensor of shape [batch_size, num_choices, sequence_length] - with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts - `extract_features.py`, `run_classifier.py` and `run_squad.py`) - `token_type_ids`: an optional torch.LongTensor of shape [batch_size, num_choices, sequence_length] - with the token types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` - and type 1 corresponds to a `sentence B` token (see BERT paper for more details). - `attention_mask`: an optional torch.LongTensor of shape [batch_size, num_choices, sequence_length] with indices - selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max - input sequence length in the current batch. It's the mask that we typically use for attention when - a batch has varying length sentences. - `labels`: labels for the classification output: torch.LongTensor of shape [batch_size] - with indices selected in [0, ..., num_choices]. - Outputs: - if `labels` is not `None`: - Outputs the CrossEntropy classification loss of the output with the labels. - if `labels` is `None`: - Outputs the classification logits of shape [batch_size, num_labels]. - Example usage: - ```python - # Already been converted into WordPiece token ids - input_ids = torch.LongTensor([[[31, 51, 99], [15, 5, 0]], [[12, 16, 42], [14, 28, 57]]]) - input_mask = torch.LongTensor([[[1, 1, 1], [1, 1, 0]],[[1,1,0], [1, 0, 0]]]) - token_type_ids = torch.LongTensor([[[0, 0, 1], [0, 1, 0]],[[0, 1, 1], [0, 0, 1]]]) - config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768, - num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072) - num_choices = 2 - model = BertForMultipleChoice(num_choices, config, bert_dir) - logits = model(input_ids, token_type_ids, input_mask) - ``` """ - def __init__(self, num_choices, config=None, bert_dir=None): + BERT model for multiple choice. + + :param fastNLP.embeddings.BertEmbedding embed: 下游模型的编码器(encoder). + :param int num_choices: 多选任务选项数目，默认值为2. + :param float dropout: dropout的大小，默认值为0.1. + """ + def __init__(self, embed: BertEmbedding, num_choices=2, dropout=0.1): super(BertForMultipleChoice, self).__init__() + self.num_choices = num_choices - if bert_dir is not None: - self.bert = BertModel.from_pretrained(bert_dir) - else: - if config is None: - config = BertConfig(30522) - self.bert = BertModel(config) - self.dropout = nn.Dropout(config.hidden_dropout_prob) - self.classifier = nn.Linear(config.hidden_size, 1) - - @classmethod - def from_pretrained(cls, num_choices, pretrained_model_dir): - config = BertConfig(pretrained_model_dir) - model = cls(num_choices=num_choices, config=config, bert_dir=pretrained_model_dir) - return model - - def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None): - flat_input_ids = input_ids.view(-1, input_ids.size(-1)) - flat_token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) - flat_attention_mask = attention_mask.view(-1, attention_mask.size(-1)) - _, pooled_output = self.bert(flat_input_ids, flat_token_type_ids, flat_attention_mask, output_all_encoded_layers=False) - pooled_output = self.dropout(pooled_output) + self.bert = embed + self.dropout = nn.Dropout(p=dropout) + self.classifier = nn.Linear(self.bert.embedding_dim, 1) + + if not self.bert.model.include_cls_sep: + self.bert.model.include_cls_sep = True + warn_msg = "Bert for multiple choice excepts BertEmbedding `include_cls_sep` True, " \ + "but got False. FastNLP has changed it to True." + logger.warn(warn_msg) + warnings.warn(warn_msg) + + def forward(self, words): + """ + :param torch.LongTensor words: [batch_size, num_choices, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.LongTensor [batch_size, num_choices] + """ + batch_size, num_choices, seq_len = words.size() + + input_ids = words.view(batch_size * num_choices, seq_len) + hidden = self.bert(input_ids) + pooled_output = self.dropout(hidden[:, 0]) logits = self.classifier(pooled_output) reshaped_logits = logits.view(-1, self.num_choices) - if labels is not None: - loss_fct = nn.CrossEntropyLoss() - loss = loss_fct(reshaped_logits, labels) - return {Const.OUTPUT: reshaped_logits, Const.LOSS: loss} - else: - return {Const.OUTPUT: reshaped_logits} + return {Const.OUTPUT: reshaped_logits} - def predict(self, input_ids, token_type_ids=None, attention_mask=None): - logits = self.forward(input_ids, token_type_ids, attention_mask)[Const.OUTPUT] + def predict(self, words): + """ + :param torch.LongTensor words: [batch_size, num_choices, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.LongTensor [batch_size] + """ + logits = self.forward(words)[Const.OUTPUT] return {Const.OUTPUT: torch.argmax(logits, dim=-1)} class BertForTokenClassification(BaseModel): - """BERT model for token-level classification. - This module is composed of the BERT model with a linear layer on top of - the full hidden state of the last layer. - Params: - `config`: a BertConfig class instance with the configuration to build a new model. - `num_labels`: the number of classes for the classifier. Default = 2. - `bert_dir`: a dir which contains the bert parameters within file `pytorch_model.bin` - Inputs: - `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length] - with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts - `extract_features.py`, `run_classifier.py` and `run_squad.py`) - `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token - types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to - a `sentence B` token (see BERT paper for more details). - `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices - selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max - input sequence length in the current batch. It's the mask that we typically use for attention when - a batch has varying length sentences. - `labels`: labels for the classification output: torch.LongTensor of shape [batch_size, sequence_length] - with indices selected in [0, ..., num_labels]. - Outputs: - if `labels` is not `None`: - Outputs the CrossEntropy classification loss of the output with the labels. - if `labels` is `None`: - Outputs the classification logits of shape [batch_size, sequence_length, num_labels]. - Example usage: - ```python - # Already been converted into WordPiece token ids - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) - config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768, - num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072) - num_labels = 2 - bert_dir = 'your-bert-file-dir' - model = BertForTokenClassification(num_labels, config, bert_dir) - logits = model(input_ids, token_type_ids, input_mask) - ``` """ - def __init__(self, num_labels, config=None, bert_dir=None): + BERT model for token classification. + + :param fastNLP.embeddings.BertEmbedding embed: 下游模型的编码器(encoder). + :param int num_labels: 序列标注标签数目，无默认值. + :param float dropout: dropout的大小，默认值为0.1. + """ + def __init__(self, embed: BertEmbedding, num_labels, dropout=0.1): super(BertForTokenClassification, self).__init__() + self.num_labels = num_labels - if bert_dir is not None: - self.bert = BertModel.from_pretrained(bert_dir) - else: - if config is None: - config = BertConfig(30522) - self.bert = BertModel(config) - self.dropout = nn.Dropout(config.hidden_dropout_prob) - self.classifier = nn.Linear(config.hidden_size, num_labels) - - @classmethod - def from_pretrained(cls, num_labels, pretrained_model_dir): - config = BertConfig(pretrained_model_dir) - model = cls(num_labels=num_labels, config=config, bert_dir=pretrained_model_dir) - return model - - def forward(self, input_ids, token_type_ids=None, attention_mask=None, labels=None): - sequence_output, _ = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False) + self.bert = embed + self.dropout = nn.Dropout(p=dropout) + self.classifier = nn.Linear(self.bert.embedding_dim, num_labels) + + if self.bert.model.include_cls_sep: + self.bert.model.include_cls_sep = False + warn_msg = "Bert for token classification excepts BertEmbedding `include_cls_sep` False, " \ + "but got True. FastNLP has changed it to False." + logger.warn(warn_msg) + warnings.warn(warn_msg) + + def forward(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.Tensor [batch_size, seq_len, num_labels] + """ + sequence_output = self.bert(words) # [batch_size, seq_len, embed_dim] sequence_output = self.dropout(sequence_output) logits = self.classifier(sequence_output) - if labels is not None: - loss_fct = nn.CrossEntropyLoss() - # Only keep active parts of the loss - if attention_mask is not None: - active_loss = attention_mask.view(-1) == 1 - active_logits = logits.view(-1, self.num_labels)[active_loss] - active_labels = labels.view(-1)[active_loss] - loss = loss_fct(active_logits, active_labels) - else: - loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1)) - return {Const.OUTPUT: logits, Const.LOSS: loss} - else: - return {Const.OUTPUT: logits} - - def predict(self, input_ids, token_type_ids=None, attention_mask=None): - logits = self.forward(input_ids, token_type_ids, attention_mask)[Const.OUTPUT] + return {Const.OUTPUT: logits} + + def predict(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: { :attr:`fastNLP.Const.OUTPUT` : logits}: torch.LongTensor [batch_size, seq_len] + """ + logits = self.forward(words)[Const.OUTPUT] return {Const.OUTPUT: torch.argmax(logits, dim=-1)} class BertForQuestionAnswering(BaseModel): - """BERT model for Question Answering (span extraction). - This module is composed of the BERT model with a linear layer on top of - the sequence output that computes start_logits and end_logits - Params: - `config`: a BertConfig class instance with the configuration to build a new model. - `bert_dir`: a dir which contains the bert parameters within file `pytorch_model.bin` - Inputs: - `input_ids`: a torch.LongTensor of shape [batch_size, sequence_length] - with the word token indices in the vocabulary(see the tokens preprocessing logic in the scripts - `extract_features.py`, `run_classifier.py` and `run_squad.py`) - `token_type_ids`: an optional torch.LongTensor of shape [batch_size, sequence_length] with the token - types indices selected in [0, 1]. Type 0 corresponds to a `sentence A` and type 1 corresponds to - a `sentence B` token (see BERT paper for more details). - `attention_mask`: an optional torch.LongTensor of shape [batch_size, sequence_length] with indices - selected in [0, 1]. It's a mask to be used if the input sequence length is smaller than the max - input sequence length in the current batch. It's the mask that we typically use for attention when - a batch has varying length sentences. - `start_positions`: position of the first token for the labeled span: torch.LongTensor of shape [batch_size]. - Positions are clamped to the length of the sequence and position outside of the sequence are not taken - into account for computing the loss. - `end_positions`: position of the last token for the labeled span: torch.LongTensor of shape [batch_size]. - Positions are clamped to the length of the sequence and position outside of the sequence are not taken - into account for computing the loss. - Outputs: - if `start_positions` and `end_positions` are not `None`: - Outputs the total_loss which is the sum of the CrossEntropy loss for the start and end token positions. - if `start_positions` or `end_positions` is `None`: - Outputs a tuple of start_logits, end_logits which are the logits respectively for the start and end - position tokens of shape [batch_size, sequence_length]. - Example usage: - ```python - # Already been converted into WordPiece token ids - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) - config = BertConfig(vocab_size_or_config_json_file=32000, hidden_size=768, - num_hidden_layers=12, num_attention_heads=12, intermediate_size=3072) - bert_dir = 'your-bert-file-dir' - model = BertForQuestionAnswering(config, bert_dir) - start_logits, end_logits = model(input_ids, token_type_ids, input_mask) - ``` """ - def __init__(self, config=None, bert_dir=None): + BERT model for classification. + + :param fastNLP.embeddings.BertEmbedding embed: 下游模型的编码器(encoder). + :param int num_labels: 抽取式QA列数，默认值为2(即第一列为start_span, 第二列为end_span). + """ + def __init__(self, embed: BertEmbedding, num_labels=2): super(BertForQuestionAnswering, self).__init__() - if bert_dir is not None: - self.bert = BertModel.from_pretrained(bert_dir) - else: - if config is None: - config = BertConfig(30522) - self.bert = BertModel(config) - # TODO check with Google if it's normal there is no dropout on the token classifier of SQuAD in the TF version - # self.dropout = nn.Dropout(config.hidden_dropout_prob) - self.qa_outputs = nn.Linear(config.hidden_size, 2) - - @classmethod - def from_pretrained(cls, pretrained_model_dir): - config = BertConfig(pretrained_model_dir) - model = cls(config=config, bert_dir=pretrained_model_dir) - return model - - def forward(self, input_ids, token_type_ids=None, attention_mask=None, start_positions=None, end_positions=None): - sequence_output, _ = self.bert(input_ids, token_type_ids, attention_mask, output_all_encoded_layers=False) - logits = self.qa_outputs(sequence_output) - start_logits, end_logits = logits.split(1, dim=-1) - start_logits = start_logits.squeeze(-1) - end_logits = end_logits.squeeze(-1) - - if start_positions is not None and end_positions is not None: - # If we are on multi-GPU, split add a dimension - if len(start_positions.size()) > 1: - start_positions = start_positions.squeeze(-1) - if len(end_positions.size()) > 1: - end_positions = end_positions.squeeze(-1) - # sometimes the start/end positions are outside our model inputs, we ignore these terms - ignored_index = start_logits.size(1) - start_positions.clamp_(0, ignored_index) - end_positions.clamp_(0, ignored_index) - - loss_fct = nn.CrossEntropyLoss(ignore_index=ignored_index) - start_loss = loss_fct(start_logits, start_positions) - end_loss = loss_fct(end_logits, end_positions) - total_loss = (start_loss + end_loss) / 2 - return {Const.OUTPUTS(0): start_logits, Const.OUTPUTS(1): end_logits, Const.LOSS: total_loss} - else: - return {Const.OUTPUTS(0): start_logits, Const.OUTPUTS(1): end_logits} - - def predict(self, input_ids, token_type_ids=None, attention_mask=None, **kwargs): - logits = self.forward(input_ids, token_type_ids, attention_mask) - start_logits = logits[Const.OUTPUTS(0)] - end_logits = logits[Const.OUTPUTS(1)] - return {Const.OUTPUTS(0): torch.argmax(start_logits, dim=-1), - Const.OUTPUTS(1): torch.argmax(end_logits, dim=-1)} + + self.bert = embed + self.num_labels = num_labels + self.qa_outputs = nn.Linear(self.bert.embedding_dim, self.num_labels) + + if not self.bert.model.include_cls_sep: + self.bert.model.include_cls_sep = True + warn_msg = "Bert for question answering excepts BertEmbedding `include_cls_sep` True, " \ + "but got False. FastNLP has changed it to True." + logger.warn(warn_msg) + warnings.warn(warn_msg) + + def forward(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: 一个包含num_labels个logit的dict，每一个logit的形状都是[batch_size, seq_len + 2] + """ + sequence_output = self.bert(words) + logits = self.qa_outputs(sequence_output) # [batch_size, seq_len, num_labels] + + return {Const.OUTPUTS(i): logits[:, :, i] for i in range(self.num_labels)} + + def predict(self, words): + """ + :param torch.LongTensor words: [batch_size, seq_len] + :return: 一个包含num_labels个logit的dict，每一个logit的形状都是[batch_size] + """ + logits = self.forward(words) + return {Const.OUTPUTS(i): torch.argmax(logits[Const.OUTPUTS(i)], dim=-1) for i in range(self.num_labels)} diff --git a/fastNLP/models/biaffine_parser.py b/fastNLP/models/biaffine_parser.py index 29487864..5d094472 100644 --- a/fastNLP/models/biaffine_parser.py +++ b/fastNLP/models/biaffine_parser.py @@ -130,8 +130,6 @@ def _find_cycle(vertices, edges): class GraphParser(BaseModel): """ - 别名：:class:`fastNLP.models.GraphParser` :class:`fastNLP.models.baffine_parser.GraphParser` - 基于图的parser base class, 支持贪婪解码和最大生成树解码 """ @@ -150,7 +148,7 @@ class GraphParser(BaseModel): """ _, seq_len, _ = arc_matrix.shape matrix = arc_matrix + torch.diag(arc_matrix.new(seq_len).fill_(-np.inf)) - flip_mask = (mask == 0).byte() + flip_mask = mask.eq(0) matrix.masked_fill_(flip_mask.unsqueeze(1), -np.inf) _, heads = torch.max(matrix, dim=2) if mask is not None: @@ -207,7 +205,7 @@ class ArcBiaffine(nn.Module): output = dep.matmul(self.U) output = output.bmm(head.transpose(-1, -2)) if self.has_bias: - output += head.matmul(self.bias).unsqueeze(1) + output = output + head.matmul(self.bias).unsqueeze(1) return output @@ -234,18 +232,16 @@ class LabelBilinear(nn.Module): :return output: [batch, seq_len, num_cls] 每个元素对应类别的概率图 """ output = self.bilinear(x1, x2) - output += self.lin(torch.cat([x1, x2], dim=2)) + output = output + self.lin(torch.cat([x1, x2], dim=2)) return output class BiaffineParser(GraphParser): """ - 别名：:class:`fastNLP.models.BiaffineParser` :class:`fastNLP.models.baffine_parser.BiaffineParser` - Biaffine Dependency Parser 实现. 论文参考 `Deep Biaffine Attention for Neural Dependency Parsing (Dozat and Manning, 2016) `_ . - :param init_embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 + :param embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 embedding的大小和每个词的维度. 也可以传入 nn.Embedding 对象, 此时就以传入的对象作为embedding :param pos_vocab_size: part-of-speech 词典大小 @@ -262,7 +258,7 @@ class BiaffineParser(GraphParser): """ def __init__(self, - init_embed, + embed, pos_vocab_size, pos_emb_dim, num_label, @@ -276,7 +272,7 @@ class BiaffineParser(GraphParser): super(BiaffineParser, self).__init__() rnn_out_size = 2 * rnn_hidden_size word_hid_dim = pos_hid_dim = rnn_hidden_size - self.word_embedding = get_embeddings(init_embed) + self.word_embedding = get_embeddings(embed) word_emb_dim = self.word_embedding.embedding_dim self.pos_embedding = nn.Embedding(num_embeddings=pos_vocab_size, embedding_dim=pos_emb_dim) self.word_fc = nn.Linear(word_emb_dim, word_hid_dim) @@ -363,7 +359,7 @@ class BiaffineParser(GraphParser): # print('forward {} {}'.format(batch_size, seq_len)) # get sequence mask - mask = seq_len_to_mask(seq_len).long() + mask = seq_len_to_mask(seq_len, max_len=length).long() word = self.word_embedding(words1) # [N,L] -> [N,L,C_0] pos = self.pos_embedding(words2) # [N,L] -> [N,L,C_1] @@ -435,10 +431,10 @@ class BiaffineParser(GraphParser): """ batch_size, length, _ = pred1.shape - mask = seq_len_to_mask(seq_len) + mask = seq_len_to_mask(seq_len, max_len=length) flip_mask = (mask == 0) _arc_pred = pred1.clone() - _arc_pred.masked_fill_(flip_mask.unsqueeze(1), -float('inf')) + _arc_pred = _arc_pred.masked_fill(flip_mask.unsqueeze(1), -float('inf')) arc_logits = F.log_softmax(_arc_pred, dim=2) label_logits = F.log_softmax(pred2, dim=2) batch_index = torch.arange(batch_size, device=arc_logits.device, dtype=torch.long).unsqueeze(1) @@ -446,9 +442,8 @@ class BiaffineParser(GraphParser): arc_loss = arc_logits[batch_index, child_index, target1] label_loss = label_logits[batch_index, child_index, target2] - byte_mask = flip_mask.byte() - arc_loss.masked_fill_(byte_mask, 0) - label_loss.masked_fill_(byte_mask, 0) + arc_loss = arc_loss.masked_fill(flip_mask, 0) + label_loss = label_loss.masked_fill(flip_mask, 0) arc_nll = -arc_loss.mean() label_nll = -label_loss.mean() return arc_nll + label_nll @@ -476,8 +471,6 @@ class BiaffineParser(GraphParser): class ParserLoss(LossFunc): """ - 别名：:class:`fastNLP.models.ParserLoss` :class:`fastNLP.models.baffine_parser.ParserLoss` - 计算parser的loss :param pred1: [batch_size, seq_len, seq_len] 边预测logits @@ -501,8 +494,6 @@ class ParserLoss(LossFunc): class ParserMetric(MetricBase): """ - 别名：:class:`fastNLP.models.ParserMetric` :class:`fastNLP.models.baffine_parser.ParserMetric` - 评估parser的性能 :param pred1: 边预测logits diff --git a/fastNLP/models/cnn_text_classification.py b/fastNLP/models/cnn_text_classification.py index e00a0697..65c20a55 100644 --- a/fastNLP/models/cnn_text_classification.py +++ b/fastNLP/models/cnn_text_classification.py @@ -1,3 +1,8 @@ +""" +.. todo:: + doc +""" + __all__ = [ "CNNText" ] @@ -7,18 +12,16 @@ import torch.nn as nn from ..core.const import Const as C from ..core.utils import seq_len_to_mask -from ..modules import encoder from ..embeddings import embedding +from ..modules import encoder class CNNText(torch.nn.Module): """ - 别名：:class:`fastNLP.models.CNNText` :class:`fastNLP.models.cnn_text_classification.CNNText` - 使用CNN进行文本分类的模型 'Yoon Kim. 2014. Convolution Neural Networks for Sentence Classification.' - :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray init_embed: Embedding的大小(传入tuple(int, int), + :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray embed: Embedding的大小(传入tuple(int, int), 第一个int为vocab_zie, 第二个int为embed_dim); 如果为Tensor, Embedding, ndarray等则直接使用该值初始化Embedding :param int num_classes: 一共有多少类 :param int,tuple(int) out_channels: 输出channel的数量。如果为list，则需要与kernel_sizes的数量保持一致 @@ -26,7 +29,7 @@ class CNNText(torch.nn.Module): :param float dropout: Dropout的大小 """ - def __init__(self, init_embed, + def __init__(self, embed, num_classes, kernel_nums=(30, 40, 50), kernel_sizes=(1, 3, 5), @@ -34,7 +37,7 @@ class CNNText(torch.nn.Module): super(CNNText, self).__init__() # no support for pre-trained embedding currently - self.embed = embedding.Embedding(init_embed) + self.embed = embedding.Embedding(embed) self.conv_pool = encoder.ConvMaxpool( in_channels=self.embed.embedding_dim, out_channels=kernel_nums, diff --git a/fastNLP/models/enas_controller.py b/fastNLP/models/enas_controller.py deleted file mode 100644 index e83c6b51..00000000 --- a/fastNLP/models/enas_controller.py +++ /dev/null @@ -1,223 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch -"""A module with NAS controller-related code.""" -import collections -import os - -import torch -import torch.nn.functional as F - -from . import enas_utils as utils -from .enas_utils import Node - - -def _construct_dags(prev_nodes, activations, func_names, num_blocks): - """Constructs a set of DAGs based on the actions, i.e., previous nodes and - activation functions, sampled from the controller/policy pi. - - Args: - prev_nodes: Previous node actions from the policy. - activations: Activations sampled from the policy. - func_names: Mapping from activation function names to functions. - num_blocks: Number of blocks in the target RNN cell. - - Returns: - A list of DAGs defined by the inputs. - - RNN cell DAGs are represented in the following way: - - 1. Each element (node) in a DAG is a list of `Node`s. - - 2. The `Node`s in the list dag[i] correspond to the subsequent nodes - that take the output from node i as their own input. - - 3. dag[-1] is the node that takes input from x^{(t)} and h^{(t - 1)}. - dag[-1] always feeds dag[0]. - dag[-1] acts as if `w_xc`, `w_hc`, `w_xh` and `w_hh` are its - weights. - - 4. dag[N - 1] is the node that produces the hidden state passed to - the next timestep. dag[N - 1] is also always a leaf node, and therefore - is always averaged with the other leaf nodes and fed to the output - decoder. - """ - dags = [] - for nodes, func_ids in zip(prev_nodes, activations): - dag = collections.defaultdict(list) - - # add first node - dag[-1] = [Node(0, func_names[func_ids[0]])] - dag[-2] = [Node(0, func_names[func_ids[0]])] - - # add following nodes - for jdx, (idx, func_id) in enumerate(zip(nodes, func_ids[1:])): - dag[utils.to_item(idx)].append(Node(jdx + 1, func_names[func_id])) - - leaf_nodes = set(range(num_blocks)) - dag.keys() - - # merge with avg - for idx in leaf_nodes: - dag[idx] = [Node(num_blocks, 'avg')] - - # This is actually y^{(t)}. h^{(t)} is node N - 1 in - # the graph, where N Is the number of nodes. I.e., h^{(t)} takes - # only one other node as its input. - # last h[t] node - last_node = Node(num_blocks + 1, 'h[t]') - dag[num_blocks] = [last_node] - dags.append(dag) - - return dags - - -class Controller(torch.nn.Module): - """Based on - https://github.com/pytorch/examples/blob/master/word_language_model/model.py - - RL controllers do not necessarily have much to do with - language models. - - Base the controller RNN on the GRU from: - https://github.com/ikostrikov/pytorch-a2c-ppo-acktr/blob/master/model.py - """ - def __init__(self, num_blocks=4, controller_hid=100, cuda=False): - torch.nn.Module.__init__(self) - - # `num_tokens` here is just the activation function - # for every even step, - self.shared_rnn_activations = ['tanh', 'ReLU', 'identity', 'sigmoid'] - self.num_tokens = [len(self.shared_rnn_activations)] - self.controller_hid = controller_hid - self.use_cuda = cuda - self.num_blocks = num_blocks - for idx in range(num_blocks): - self.num_tokens += [idx + 1, len(self.shared_rnn_activations)] - self.func_names = self.shared_rnn_activations - - num_total_tokens = sum(self.num_tokens) - - self.encoder = torch.nn.Embedding(num_total_tokens, - controller_hid) - self.lstm = torch.nn.LSTMCell(controller_hid, controller_hid) - - # Perhaps these weights in the decoder should be - # shared? At least for the activation functions, which all have the - # same size. - self.decoders = [] - for idx, size in enumerate(self.num_tokens): - decoder = torch.nn.Linear(controller_hid, size) - self.decoders.append(decoder) - - self._decoders = torch.nn.ModuleList(self.decoders) - - self.reset_parameters() - self.static_init_hidden = utils.keydefaultdict(self.init_hidden) - - def _get_default_hidden(key): - return utils.get_variable( - torch.zeros(key, self.controller_hid), - self.use_cuda, - requires_grad=False) - - self.static_inputs = utils.keydefaultdict(_get_default_hidden) - - def reset_parameters(self): - init_range = 0.1 - for param in self.parameters(): - param.data.uniform_(-init_range, init_range) - for decoder in self.decoders: - decoder.bias.data.fill_(0) - - def forward(self, # pylint:disable=arguments-differ - inputs, - hidden, - block_idx, - is_embed): - if not is_embed: - embed = self.encoder(inputs) - else: - embed = inputs - - hx, cx = self.lstm(embed, hidden) - logits = self.decoders[block_idx](hx) - - logits /= 5.0 - - # # exploration - # if self.args.mode == 'train': - # logits = (2.5 * F.tanh(logits)) - - return logits, (hx, cx) - - def sample(self, batch_size=1, with_details=False, save_dir=None): - """Samples a set of `args.num_blocks` many computational nodes from the - controller, where each node is made up of an activation function, and - each node except the last also includes a previous node. - """ - if batch_size < 1: - raise Exception(f'Wrong batch_size: {batch_size} < 1') - - # [B, L, H] - inputs = self.static_inputs[batch_size] - hidden = self.static_init_hidden[batch_size] - - activations = [] - entropies = [] - log_probs = [] - prev_nodes = [] - # The RNN controller alternately outputs an activation, - # followed by a previous node, for each block except the last one, - # which only gets an activation function. The last node is the output - # node, and its previous node is the average of all leaf nodes. - for block_idx in range(2*(self.num_blocks - 1) + 1): - logits, hidden = self.forward(inputs, - hidden, - block_idx, - is_embed=(block_idx == 0)) - - probs = F.softmax(logits, dim=-1) - log_prob = F.log_softmax(logits, dim=-1) - # .mean() for entropy? - entropy = -(log_prob * probs).sum(1, keepdim=False) - - action = probs.multinomial(num_samples=1).data - selected_log_prob = log_prob.gather( - 1, utils.get_variable(action, requires_grad=False)) - - # why the [:, 0] here? Should it be .squeeze(), or - # .view()? Same below with `action`. - entropies.append(entropy) - log_probs.append(selected_log_prob[:, 0]) - - # 0: function, 1: previous node - mode = block_idx % 2 - inputs = utils.get_variable( - action[:, 0] + sum(self.num_tokens[:mode]), - requires_grad=False) - - if mode == 0: - activations.append(action[:, 0]) - elif mode == 1: - prev_nodes.append(action[:, 0]) - - prev_nodes = torch.stack(prev_nodes).transpose(0, 1) - activations = torch.stack(activations).transpose(0, 1) - - dags = _construct_dags(prev_nodes, - activations, - self.func_names, - self.num_blocks) - - if save_dir is not None: - for idx, dag in enumerate(dags): - utils.draw_network(dag, - os.path.join(save_dir, f'graph{idx}.png')) - - if with_details: - return dags, torch.cat(log_probs), torch.cat(entropies) - - return dags - - def init_hidden(self, batch_size): - zeros = torch.zeros(batch_size, self.controller_hid) - return (utils.get_variable(zeros, self.use_cuda, requires_grad=False), - utils.get_variable(zeros.clone(), self.use_cuda, requires_grad=False)) diff --git a/fastNLP/models/enas_model.py b/fastNLP/models/enas_model.py deleted file mode 100644 index b6b683c0..00000000 --- a/fastNLP/models/enas_model.py +++ /dev/null @@ -1,390 +0,0 @@ -""" -Module containing the shared RNN model. -Code Modified from https://github.com/carpedm20/ENAS-pytorch -""" -import collections - -import numpy as np -import torch -import torch.nn as nn -import torch.nn.functional as F -from torch.autograd import Variable - -from . import enas_utils as utils -from .base_model import BaseModel - - -def _get_dropped_weights(w_raw, dropout_p, is_training): - """Drops out weights to implement DropConnect. - - Args: - w_raw: Full, pre-dropout, weights to be dropped out. - dropout_p: Proportion of weights to drop out. - is_training: True iff _shared_ model is training. - - Returns: - The dropped weights. - - Why does torch.nn.functional.dropout() return: - 1. `torch.autograd.Variable()` on the training loop - 2. `torch.nn.Parameter()` on the controller or eval loop, when - training = False... - - Even though the call to `_setweights` in the Smerity repo's - `weight_drop.py` does not have this behaviour, and `F.dropout` always - returns `torch.autograd.Variable` there, even when `training=False`? - - The above TODO is the reason for the hacky check for `torch.nn.Parameter`. - """ - dropped_w = F.dropout(w_raw, p=dropout_p, training=is_training) - - if isinstance(dropped_w, torch.nn.Parameter): - dropped_w = dropped_w.clone() - - return dropped_w - - -class EmbeddingDropout(torch.nn.Embedding): - """Class for dropping out embeddings by zero'ing out parameters in the - embedding matrix. - - This is equivalent to dropping out particular words, e.g., in the sentence - 'the quick brown fox jumps over the lazy dog', dropping out 'the' would - lead to the sentence '### quick brown fox jumps over ### lazy dog' (in the - embedding vector space). - - See 'A Theoretically Grounded Application of Dropout in Recurrent Neural - Networks', (Gal and Ghahramani, 2016). - """ - - def __init__(self, - num_embeddings, - embedding_dim, - max_norm=None, - norm_type=2, - scale_grad_by_freq=False, - sparse=False, - dropout=0.1, - scale=None): - """Embedding constructor. - - Args: - dropout: Dropout probability. - scale: Used to scale parameters of embedding weight matrix that are - not dropped out. Note that this is _in addition_ to the - `1/(1 - dropout)` scaling. - - See `torch.nn.Embedding` for remaining arguments. - """ - torch.nn.Embedding.__init__(self, - num_embeddings=num_embeddings, - embedding_dim=embedding_dim, - max_norm=max_norm, - norm_type=norm_type, - scale_grad_by_freq=scale_grad_by_freq, - sparse=sparse) - self.dropout = dropout - assert (dropout >= 0.0) and (dropout < 1.0), ('Dropout must be >= 0.0 ' - 'and < 1.0') - self.scale = scale - - def forward(self, inputs): # pylint:disable=arguments-differ - """Embeds `inputs` with the dropped out embedding weight matrix.""" - if self.training: - dropout = self.dropout - else: - dropout = 0 - - if dropout: - mask = self.weight.data.new(self.weight.size(0), 1) - mask.bernoulli_(1 - dropout) - mask = mask.expand_as(self.weight) - mask = mask / (1 - dropout) - masked_weight = self.weight * Variable(mask) - else: - masked_weight = self.weight - if self.scale and self.scale != 1: - masked_weight = masked_weight * self.scale - - return F.embedding(inputs, - masked_weight, - max_norm=self.max_norm, - norm_type=self.norm_type, - scale_grad_by_freq=self.scale_grad_by_freq, - sparse=self.sparse) - - -class LockedDropout(nn.Module): - # code from https://github.com/salesforce/awd-lstm-lm/blob/master/locked_dropout.py - def __init__(self): - super().__init__() - - def forward(self, x, dropout=0.5): - if not self.training or not dropout: - return x - m = x.data.new(1, x.size(1), x.size(2)).bernoulli_(1 - dropout) - mask = Variable(m, requires_grad=False) / (1 - dropout) - mask = mask.expand_as(x) - return mask * x - - -class ENASModel(BaseModel): - """Shared RNN model.""" - - def __init__(self, embed_num, num_classes, num_blocks=4, cuda=False, shared_hid=1000, shared_embed=1000): - super(ENASModel, self).__init__() - - self.use_cuda = cuda - - self.shared_hid = shared_hid - self.num_blocks = num_blocks - self.decoder = nn.Linear(self.shared_hid, num_classes) - self.encoder = EmbeddingDropout(embed_num, - shared_embed, - dropout=0.1) - self.lockdrop = LockedDropout() - self.dag = None - - # Tie weights - # self.decoder.weight = self.encoder.weight - - # Since W^{x, c} and W^{h, c} are always summed, there - # is no point duplicating their bias offset parameter. Likewise for - # W^{x, h} and W^{h, h}. - self.w_xc = nn.Linear(shared_embed, self.shared_hid) - self.w_xh = nn.Linear(shared_embed, self.shared_hid) - - # The raw weights are stored here because the hidden-to-hidden weights - # are weight dropped on the forward pass. - self.w_hc_raw = torch.nn.Parameter( - torch.Tensor(self.shared_hid, self.shared_hid)) - self.w_hh_raw = torch.nn.Parameter( - torch.Tensor(self.shared_hid, self.shared_hid)) - self.w_hc = None - self.w_hh = None - - self.w_h = collections.defaultdict(dict) - self.w_c = collections.defaultdict(dict) - - for idx in range(self.num_blocks): - for jdx in range(idx + 1, self.num_blocks): - self.w_h[idx][jdx] = nn.Linear(self.shared_hid, - self.shared_hid, - bias=False) - self.w_c[idx][jdx] = nn.Linear(self.shared_hid, - self.shared_hid, - bias=False) - - self._w_h = nn.ModuleList([self.w_h[idx][jdx] - for idx in self.w_h - for jdx in self.w_h[idx]]) - self._w_c = nn.ModuleList([self.w_c[idx][jdx] - for idx in self.w_c - for jdx in self.w_c[idx]]) - - self.batch_norm = None - # if args.mode == 'train': - # self.batch_norm = nn.BatchNorm1d(self.shared_hid) - # else: - # self.batch_norm = None - - self.reset_parameters() - self.static_init_hidden = utils.keydefaultdict(self.init_hidden) - - def setDAG(self, dag): - if self.dag is None: - self.dag = dag - - def forward(self, word_seq, hidden=None): - inputs = torch.transpose(word_seq, 0, 1) - - time_steps = inputs.size(0) - batch_size = inputs.size(1) - - self.w_hh = _get_dropped_weights(self.w_hh_raw, - 0.5, - self.training) - self.w_hc = _get_dropped_weights(self.w_hc_raw, - 0.5, - self.training) - - # hidden = self.static_init_hidden[batch_size] if hidden is None else hidden - hidden = self.static_init_hidden[batch_size] - - embed = self.encoder(inputs) - - embed = self.lockdrop(embed, 0.65 if self.training else 0) - - # The norm of hidden states are clipped here because - # otherwise ENAS is especially prone to exploding activations on the - # forward pass. This could probably be fixed in a more elegant way, but - # it might be exposing a weakness in the ENAS algorithm as currently - # proposed. - # - # For more details, see - # https://github.com/carpedm20/ENAS-pytorch/issues/6 - clipped_num = 0 - max_clipped_norm = 0 - h1tohT = [] - logits = [] - for step in range(time_steps): - x_t = embed[step] - logit, hidden = self.cell(x_t, hidden, self.dag) - - hidden_norms = hidden.norm(dim=-1) - max_norm = 25.0 - if hidden_norms.data.max() > max_norm: - # Just directly use the torch slice operations - # in PyTorch v0.4. - # - # This workaround for PyTorch v0.3.1 does everything in numpy, - # because the PyTorch slicing and slice assignment is too - # flaky. - hidden_norms = hidden_norms.data.cpu().numpy() - - clipped_num += 1 - if hidden_norms.max() > max_clipped_norm: - max_clipped_norm = hidden_norms.max() - - clip_select = hidden_norms > max_norm - clip_norms = hidden_norms[clip_select] - - mask = np.ones(hidden.size()) - normalizer = max_norm / clip_norms - normalizer = normalizer[:, np.newaxis] - - mask[clip_select] = normalizer - - if self.use_cuda: - hidden *= torch.autograd.Variable( - torch.FloatTensor(mask).cuda(), requires_grad=False) - else: - hidden *= torch.autograd.Variable( - torch.FloatTensor(mask), requires_grad=False) - logits.append(logit) - h1tohT.append(hidden) - - h1tohT = torch.stack(h1tohT) - output = torch.stack(logits) - raw_output = output - - output = self.lockdrop(output, 0.4 if self.training else 0) - - # Pooling - output = torch.mean(output, 0) - - decoded = self.decoder(output) - - extra_out = {'dropped': decoded, - 'hiddens': h1tohT, - 'raw': raw_output} - return {'pred': decoded, 'hidden': hidden, 'extra_out': extra_out} - - def cell(self, x, h_prev, dag): - """Computes a single pass through the discovered RNN cell.""" - c = {} - h = {} - f = {} - - f[0] = self.get_f(dag[-1][0].name) - c[0] = torch.sigmoid(self.w_xc(x) + F.linear(h_prev, self.w_hc, None)) - h[0] = (c[0] * f[0](self.w_xh(x) + F.linear(h_prev, self.w_hh, None)) + - (1 - c[0]) * h_prev) - - leaf_node_ids = [] - q = collections.deque() - q.append(0) - - # Computes connections from the parent nodes `node_id` - # to their child nodes `next_id` recursively, skipping leaf nodes. A - # leaf node is a node whose id == `self.num_blocks`. - # - # Connections between parent i and child j should be computed as - # h_j = c_j*f_{ij}{(W^h_{ij}*h_i)} + (1 - c_j)*h_i, - # where c_j = \sigmoid{(W^c_{ij}*h_i)} - # - # See Training details from Section 3.1 of the paper. - # - # The following algorithm does a breadth-first (since `q.popleft()` is - # used) search over the nodes and computes all the hidden states. - while True: - if len(q) == 0: - break - - node_id = q.popleft() - nodes = dag[node_id] - - for next_node in nodes: - next_id = next_node.id - if next_id == self.num_blocks: - leaf_node_ids.append(node_id) - assert len(nodes) == 1, ('parent of leaf node should have ' - 'only one child') - continue - - w_h = self.w_h[node_id][next_id] - w_c = self.w_c[node_id][next_id] - - f[next_id] = self.get_f(next_node.name) - c[next_id] = torch.sigmoid(w_c(h[node_id])) - h[next_id] = (c[next_id] * f[next_id](w_h(h[node_id])) + - (1 - c[next_id]) * h[node_id]) - - q.append(next_id) - - # Instead of averaging loose ends, perhaps there should - # be a set of separate unshared weights for each "loose" connection - # between each node in a cell and the output. - # - # As it stands, all weights W^h_{ij} are doing double duty by - # connecting both from i to j, as well as from i to the output. - - # average all the loose ends - leaf_nodes = [h[node_id] for node_id in leaf_node_ids] - output = torch.mean(torch.stack(leaf_nodes, 2), -1) - - # stabilizing the Updates of omega - if self.batch_norm is not None: - output = self.batch_norm(output) - - return output, h[self.num_blocks - 1] - - def init_hidden(self, batch_size): - zeros = torch.zeros(batch_size, self.shared_hid) - return utils.get_variable(zeros, self.use_cuda, requires_grad=False) - - def get_f(self, name): - name = name.lower() - if name == 'relu': - f = torch.relu - elif name == 'tanh': - f = torch.tanh - elif name == 'identity': - f = lambda x: x - elif name == 'sigmoid': - f = torch.sigmoid - return f - - @property - def num_parameters(self): - def size(p): - return np.prod(p.size()) - - return sum([size(param) for param in self.parameters()]) - - def reset_parameters(self): - init_range = 0.025 - # init_range = 0.025 if self.args.mode == 'train' else 0.04 - for param in self.parameters(): - param.data.uniform_(-init_range, init_range) - self.decoder.bias.data.fill_(0) - - def predict(self, word_seq): - """ - - :param word_seq: torch.LongTensor, [batch_size, seq_len] - :return predict: dict of torch.LongTensor, [batch_size, seq_len] - """ - output = self(word_seq) - _, predict = output['pred'].max(dim=1) - return {'pred': predict} diff --git a/fastNLP/models/enas_trainer.py b/fastNLP/models/enas_trainer.py deleted file mode 100644 index 7abcc45f..00000000 --- a/fastNLP/models/enas_trainer.py +++ /dev/null @@ -1,380 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch -import math -import numpy as np -import time -import torch - -from datetime import datetime, timedelta - -from torch.optim import Adam - -try: - from tqdm.auto import tqdm -except: - from ..core.utils import _pseudo_tqdm as tqdm - -from ..core.trainer import Trainer -from ..core.batch import DataSetIter -from ..core.callback import CallbackManager, CallbackException -from ..core.dataset import DataSet -from ..core.utils import _move_dict_value_to_device -from . import enas_utils as utils -from ..core.utils import _build_args - - -def _get_no_grad_ctx_mgr(): - """Returns a the `torch.no_grad` context manager for PyTorch version >= - 0.4, or a no-op context manager otherwise. - """ - return torch.no_grad() - - -class ENASTrainer(Trainer): - """A class to wrap training code.""" - - def __init__(self, train_data, model, controller, **kwargs): - """Constructor for training algorithm. - :param DataSet train_data: the training data - :param torch.nn.modules.module model: a PyTorch model - :param torch.nn.modules.module controller: a PyTorch model - """ - self.final_epochs = kwargs['final_epochs'] - kwargs.pop('final_epochs') - super(ENASTrainer, self).__init__(train_data, model, **kwargs) - self.controller_step = 0 - self.shared_step = 0 - self.max_length = 35 - - self.shared = model - self.controller = controller - - self.shared_optim = Adam( - self.shared.parameters(), - lr=20.0, - weight_decay=1e-7) - - self.controller_optim = Adam( - self.controller.parameters(), - lr=3.5e-4) - - def train(self, load_best_model=True): - """ - :param bool load_best_model: 该参数只有在初始化提供了dev_data的情况下有效，如果True, trainer将在返回之前重新加载dev表现 - 最好的模型参数。 - :return results: 返回一个字典类型的数据, - 内含以下内容:: - - seconds: float, 表示训练时长 - 以下三个内容只有在提供了dev_data的情况下会有。 - best_eval: Dict of Dict, 表示evaluation的结果 - best_epoch: int，在第几个epoch取得的最佳值 - best_step: int, 在第几个step(batch)更新取得的最佳值 - - """ - results = {} - if self.n_epochs <= 0: - print(f"training epoch is {self.n_epochs}, nothing was done.") - results['seconds'] = 0. - return results - try: - if torch.cuda.is_available() and "cuda" in self.device: - self.model = self.model.cuda() - self._model_device = self.model.parameters().__next__().device - self._mode(self.model, is_test=False) - - self.start_time = str(datetime.now().strftime('%Y-%m-%d-%H-%M-%S')) - start_time = time.time() - print("training epochs started " + self.start_time, flush=True) - - try: - self.callback_manager.on_train_begin() - self._train() - self.callback_manager.on_train_end() - except (CallbackException, KeyboardInterrupt) as e: - self.callback_manager.on_exception(e) - - if self.dev_data is not None: - print( - "\nIn Epoch:{}/Step:{}, got best dev performance:".format(self.best_dev_epoch, self.best_dev_step) + - self.tester._format_eval_results(self.best_dev_perf), ) - results['best_eval'] = self.best_dev_perf - results['best_epoch'] = self.best_dev_epoch - results['best_step'] = self.best_dev_step - if load_best_model: - model_name = "best_" + "_".join([self.model.__class__.__name__, self.metric_key, self.start_time]) - load_succeed = self._load_model(self.model, model_name) - if load_succeed: - print("Reloaded the best model.") - else: - print("Fail to reload best model.") - finally: - pass - results['seconds'] = round(time.time() - start_time, 2) - - return results - - def _train(self): - if not self.use_tqdm: - from fastNLP.core.utils import _pseudo_tqdm as inner_tqdm - else: - inner_tqdm = tqdm - self.step = 0 - start = time.time() - total_steps = (len(self.train_data) // self.batch_size + int( - len(self.train_data) % self.batch_size != 0)) * self.n_epochs - with inner_tqdm(total=total_steps, postfix='loss:{0:<6.5f}', leave=False, dynamic_ncols=True) as pbar: - avg_loss = 0 - data_iterator = DataSetIter(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) - for epoch in range(1, self.n_epochs + 1): - pbar.set_description_str(desc="Epoch {}/{}".format(epoch, self.n_epochs)) - last_stage = (epoch > self.n_epochs + 1 - self.final_epochs) - if epoch == self.n_epochs + 1 - self.final_epochs: - print('Entering the final stage. (Only train the selected structure)') - # early stopping - self.callback_manager.on_epoch_begin() - - # 1. Training the shared parameters omega of the child models - self.train_shared(pbar) - - # 2. Training the controller parameters theta - if not last_stage: - self.train_controller() - - if ((self.validate_every > 0 and self.step % self.validate_every == 0) or - (self.validate_every < 0 and self.step % len(data_iterator) == 0)) \ - and self.dev_data is not None: - if not last_stage: - self.derive() - eval_res = self._do_validation(epoch=epoch, step=self.step) - eval_str = "Evaluation at Epoch {}/{}. Step:{}/{}. ".format(epoch, self.n_epochs, self.step, - total_steps) + \ - self.tester._format_eval_results(eval_res) - pbar.write(eval_str) - - # lr decay; early stopping - self.callback_manager.on_epoch_end() - # =============== epochs end =================== # - pbar.close() - # ============ tqdm end ============== # - - def get_loss(self, inputs, targets, hidden, dags): - """Computes the loss for the same batch for M models. - - This amounts to an estimate of the loss, which is turned into an - estimate for the gradients of the shared model. - """ - if not isinstance(dags, list): - dags = [dags] - - loss = 0 - for dag in dags: - self.shared.setDAG(dag) - inputs = _build_args(self.shared.forward, **inputs) - inputs['hidden'] = hidden - result = self.shared(**inputs) - output, hidden, extra_out = result['pred'], result['hidden'], result['extra_out'] - - self.callback_manager.on_loss_begin(targets, result) - sample_loss = self._compute_loss(result, targets) - loss += sample_loss - - assert len(dags) == 1, 'there are multiple `hidden` for multple `dags`' - return loss, hidden, extra_out - - def train_shared(self, pbar=None, max_step=None, dag=None): - """Train the language model for 400 steps of minibatches of 64 - examples. - - Args: - max_step: Used to run extra training steps as a warm-up. - dag: If not None, is used instead of calling sample(). - - BPTT is truncated at 35 timesteps. - - For each weight update, gradients are estimated by sampling M models - from the fixed controller policy, and averaging their gradients - computed on a batch of training data. - """ - model = self.shared - model.train() - self.controller.eval() - - hidden = self.shared.init_hidden(self.batch_size) - - abs_max_grad = 0 - abs_max_hidden_norm = 0 - step = 0 - raw_total_loss = 0 - total_loss = 0 - train_idx = 0 - avg_loss = 0 - data_iterator = DataSetIter(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) - - for batch_x, batch_y in data_iterator: - _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) - indices = data_iterator.get_batch_indices() - # negative sampling; replace unknown; re-weight batch_y - self.callback_manager.on_batch_begin(batch_x, batch_y, indices) - # prediction = self._data_forward(self.model, batch_x) - - dags = self.controller.sample(1) - inputs, targets = batch_x, batch_y - # self.callback_manager.on_loss_begin(batch_y, prediction) - loss, hidden, extra_out = self.get_loss(inputs, - targets, - hidden, - dags) - hidden.detach_() - - avg_loss += loss.item() - - # Is loss NaN or inf? requires_grad = False - self.callback_manager.on_backward_begin(loss) - self._grad_backward(loss) - self.callback_manager.on_backward_end() - - self._update() - self.callback_manager.on_step_end() - - if (self.step + 1) % self.print_every == 0: - if self.use_tqdm: - print_output = "loss:{0:<6.5f}".format(avg_loss / self.print_every) - pbar.update(self.print_every) - else: - end = time.time() - diff = timedelta(seconds=round(end - start)) - print_output = "[epoch: {:>3} step: {:>4}] train loss: {:>4.6} time: {}".format( - epoch, self.step, avg_loss, diff) - pbar.set_postfix_str(print_output) - avg_loss = 0 - self.step += 1 - step += 1 - self.shared_step += 1 - self.callback_manager.on_batch_end() - # ================= mini-batch end ==================== # - - def get_reward(self, dag, entropies, hidden, valid_idx=0): - """Computes the perplexity of a single sampled model on a minibatch of - validation data. - """ - if not isinstance(entropies, np.ndarray): - entropies = entropies.data.cpu().numpy() - - data_iterator = DataSetIter(self.dev_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) - - for inputs, targets in data_iterator: - valid_loss, hidden, _ = self.get_loss(inputs, targets, hidden, dag) - valid_loss = utils.to_item(valid_loss.data) - - valid_ppl = math.exp(valid_loss) - - R = 80 / valid_ppl - - rewards = R + 1e-4 * entropies - - return rewards, hidden - - def train_controller(self): - """Fixes the shared parameters and updates the controller parameters. - - The controller is updated with a score function gradient estimator - (i.e., REINFORCE), with the reward being c/valid_ppl, where valid_ppl - is computed on a minibatch of validation data. - - A moving average baseline is used. - - The controller is trained for 2000 steps per epoch (i.e., - first (Train Shared) phase -> second (Train Controller) phase). - """ - model = self.controller - model.train() - # Why can't we call shared.eval() here? Leads to loss - # being uniformly zero for the controller. - # self.shared.eval() - - avg_reward_base = None - baseline = None - adv_history = [] - entropy_history = [] - reward_history = [] - - hidden = self.shared.init_hidden(self.batch_size) - total_loss = 0 - valid_idx = 0 - for step in range(20): - # sample models - dags, log_probs, entropies = self.controller.sample( - with_details=True) - - # calculate reward - np_entropies = entropies.data.cpu().numpy() - # No gradients should be backpropagated to the - # shared model during controller training, obviously. - with _get_no_grad_ctx_mgr(): - rewards, hidden = self.get_reward(dags, - np_entropies, - hidden, - valid_idx) - - reward_history.extend(rewards) - entropy_history.extend(np_entropies) - - # moving average baseline - if baseline is None: - baseline = rewards - else: - decay = 0.95 - baseline = decay * baseline + (1 - decay) * rewards - - adv = rewards - baseline - adv_history.extend(adv) - - # policy loss - loss = -log_probs * utils.get_variable(adv, - 'cuda' in self.device, - requires_grad=False) - - loss = loss.sum() # or loss.mean() - - # update - self.controller_optim.zero_grad() - loss.backward() - - self.controller_optim.step() - - total_loss += utils.to_item(loss.data) - - if ((step % 50) == 0) and (step > 0): - reward_history, adv_history, entropy_history = [], [], [] - total_loss = 0 - - self.controller_step += 1 - # prev_valid_idx = valid_idx - # valid_idx = ((valid_idx + self.max_length) % - # (self.valid_data.size(0) - 1)) - # # Whenever we wrap around to the beginning of the - # # validation data, we reset the hidden states. - # if prev_valid_idx > valid_idx: - # hidden = self.shared.init_hidden(self.batch_size) - - def derive(self, sample_num=10, valid_idx=0): - """We are always deriving based on the very first batch - of validation data? This seems wrong... - """ - hidden = self.shared.init_hidden(self.batch_size) - - dags, _, entropies = self.controller.sample(sample_num, - with_details=True) - - max_R = 0 - best_dag = None - for dag in dags: - R, _ = self.get_reward(dag, entropies, hidden, valid_idx) - if R.max() > max_R: - max_R = R.max() - best_dag = dag - - self.model.setDAG(best_dag) diff --git a/fastNLP/models/enas_utils.py b/fastNLP/models/enas_utils.py deleted file mode 100644 index 4e402a9a..00000000 --- a/fastNLP/models/enas_utils.py +++ /dev/null @@ -1,54 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch - -from collections import defaultdict -import collections - -import numpy as np -import torch -from torch.autograd import Variable - - -def detach(h): - if type(h) == Variable: - return Variable(h.data) - else: - return tuple(detach(v) for v in h) - - -def get_variable(inputs, cuda=False, **kwargs): - if type(inputs) in [list, np.ndarray]: - inputs = torch.Tensor(inputs) - if cuda: - out = Variable(inputs.cuda(), **kwargs) - else: - out = Variable(inputs, **kwargs) - return out - - -def update_lr(optimizer, lr): - for param_group in optimizer.param_groups: - param_group['lr'] = lr - - -Node = collections.namedtuple('Node', ['id', 'name']) - - -class keydefaultdict(defaultdict): - def __missing__(self, key): - if self.default_factory is None: - raise KeyError(key) - else: - ret = self[key] = self.default_factory(key) - return ret - - -def to_item(x): - """Converts x, possibly scalar and possibly tensor, to a Python scalar.""" - if isinstance(x, (float, int)): - return x - - if float(torch.__version__[0:3]) < 0.4: - assert (x.dim() == 1) and (len(x) == 1) - return x[0] - - return x.item() diff --git a/fastNLP/models/sequence_labeling.py b/fastNLP/models/sequence_labeling.py index 4bf3f95f..d5bc250b 100644 --- a/fastNLP/models/sequence_labeling.py +++ b/fastNLP/models/sequence_labeling.py @@ -1,10 +1,10 @@ """ - 本模块实现了几种序列标注模型 +本模块实现了几种序列标注模型 """ __all__ = [ "SeqLabeling", "AdvSeqLabel", - # "BiLSTMCRF" + "BiLSTMCRF" ] import torch @@ -12,30 +12,26 @@ import torch.nn as nn import torch.nn.functional as F from .base_model import BaseModel -from ..embeddings import embedding -from ..modules import decoder, encoder -from ..modules.decoder.crf import allowed_transitions -from ..core.utils import seq_len_to_mask from ..core.const import Const as C -from ..modules import LSTM +from ..core.utils import seq_len_to_mask from ..embeddings import get_embeddings from ..modules import ConditionalRandomField +from ..modules import LSTM +from ..modules import decoder, encoder +from ..modules.decoder.crf import allowed_transitions class BiLSTMCRF(BaseModel): """ - 结构为BiLSTM + FC + Dropout + CRF. - - .. todo:: - 继续补充文档 + 结构为embedding + BiLSTM + FC + Dropout + CRF. - :param embed: tuple: - :param num_classes: - :param num_layers: - :param hidden_size: - :param dropout: - :param target_vocab: - :param encoding_type: + :param embed: 支持(1)fastNLP的各种Embedding, (2) tuple, 指明num_embedding, dimension, 如(1000, 100) + :param num_classes: 一共多少个类 + :param num_layers: BiLSTM的层数 + :param hidden_size: BiLSTM的hidden_size，实际hidden size为该值的两倍(前向、后向) + :param dropout: dropout的概率，0为不dropout + :param target_vocab: Vocabulary对象，target与index的对应关系 + :param encoding_type: encoding的类型，支持'bioes', 'bmes', 'bio', 'bmeso'等 """ def __init__(self, embed, num_classes, num_layers=1, hidden_size=100, dropout=0.5, target_vocab=None, encoding_type=None): @@ -43,14 +39,14 @@ class BiLSTMCRF(BaseModel): self.embed = get_embeddings(embed) if num_layers>1: - self.lstm = LSTM(embed.embedding_dim, num_layers=num_layers, hidden_size=hidden_size, bidirectional=True, + self.lstm = LSTM(self.embed.embedding_dim, num_layers=num_layers, hidden_size=hidden_size, bidirectional=True, batch_first=True, dropout=dropout) else: - self.lstm = LSTM(embed.embedding_dim, num_layers=num_layers, hidden_size=hidden_size, bidirectional=True, + self.lstm = LSTM(self.embed.embedding_dim, num_layers=num_layers, hidden_size=hidden_size, bidirectional=True, batch_first=True) self.dropout = nn.Dropout(dropout) - self.fc = nn.Linear(hidden_size, num_classes) + self.fc = nn.Linear(hidden_size*2, num_classes) trans = None if target_vocab is not None and encoding_type is not None: @@ -60,7 +56,7 @@ class BiLSTMCRF(BaseModel): def _forward(self, words, seq_len=None, target=None): words = self.embed(words) - feats = self.lstm(words, seq_len=seq_len) + feats, _ = self.lstm(words, seq_len=seq_len) feats = self.fc(feats) feats = self.dropout(feats) logits = F.log_softmax(feats, dim=-1) @@ -81,26 +77,23 @@ class BiLSTMCRF(BaseModel): class SeqLabeling(BaseModel): """ - 别名：:class:`fastNLP.models.SeqLabeling` :class:`fastNLP.models.sequence_labeling.SeqLabeling` - 一个基础的Sequence labeling的模型。 用于做sequence labeling的基础类。结构包含一层Embedding，一层LSTM(单向，一层)，一层FC，以及一层CRF。 - :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray init_embed: Embedding的大小(传入tuple(int, int), - 第一个int为vocab_zie, 第二个int为embed_dim); 如果为Tensor, Embedding, ndarray等则直接使用该值初始化Embedding + :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray embed: Embedding的大小(传入tuple(int, int), + 第一个int为vocab_zie, 第二个int为embed_dim); 如果为Tensor, embedding, ndarray等则直接使用该值初始化Embedding :param int hidden_size: LSTM隐藏层的大小 :param int num_classes: 一共有多少类 """ - def __init__(self, init_embed, hidden_size, num_classes): + def __init__(self, embed, hidden_size, num_classes): super(SeqLabeling, self).__init__() - self.Embedding = embedding.Embedding(init_embed) - self.Rnn = encoder.LSTM(self.Embedding.embedding_dim, hidden_size) - self.Linear = nn.Linear(hidden_size, num_classes) - self.Crf = decoder.ConditionalRandomField(num_classes) - self.mask = None - + self.embedding = get_embeddings(embed) + self.rnn = encoder.LSTM(self.embedding.embedding_dim, hidden_size) + self.fc = nn.Linear(hidden_size, num_classes) + self.crf = decoder.ConditionalRandomField(num_classes) + def forward(self, words, seq_len, target): """ :param torch.LongTensor words: [batch_size, max_len]，序列的index @@ -109,17 +102,14 @@ class SeqLabeling(BaseModel): :return y: If truth is None, return list of [decode path(list)]. Used in testing and predicting. If truth is not None, return loss, a scalar. Used in training. """ - assert words.shape[0] == seq_len.shape[0] - assert target.shape == words.shape - self.mask = self._make_mask(words, seq_len) - - x = self.Embedding(words) + mask = seq_len_to_mask(seq_len, max_len=words.size(1)) + x = self.embedding(words) # [batch_size, max_len, word_emb_dim] - x, _ = self.Rnn(x, seq_len) + x, _ = self.rnn(x, seq_len) # [batch_size, max_len, hidden_size * direction] - x = self.Linear(x) + x = self.fc(x) # [batch_size, max_len, num_classes] - return {C.LOSS: self._internal_loss(x, target)} + return {C.LOSS: self._internal_loss(x, target, mask)} def predict(self, words, seq_len): """ @@ -129,18 +119,18 @@ class SeqLabeling(BaseModel): :param torch.LongTensor seq_len: [batch_size,] :return: {'pred': xx}, [batch_size, max_len] """ - self.mask = self._make_mask(words, seq_len) + mask = seq_len_to_mask(seq_len, max_len=words.size(1)) - x = self.Embedding(words) + x = self.embedding(words) # [batch_size, max_len, word_emb_dim] - x, _ = self.Rnn(x, seq_len) + x, _ = self.rnn(x, seq_len) # [batch_size, max_len, hidden_size * direction] - x = self.Linear(x) + x = self.fc(x) # [batch_size, max_len, num_classes] - pred = self._decode(x) + pred = self._decode(x, mask) return {C.OUTPUT: pred} - def _internal_loss(self, x, y): + def _internal_loss(self, x, y, mask): """ Negative log likelihood loss. :param x: Tensor, [batch_size, max_len, tag_size] @@ -150,34 +140,23 @@ class SeqLabeling(BaseModel): """ x = x.float() y = y.long() - assert x.shape[:2] == y.shape - assert y.shape == self.mask.shape - total_loss = self.Crf(x, y, self.mask) + total_loss = self.crf(x, y, mask) return torch.mean(total_loss) - def _make_mask(self, x, seq_len): - batch_size, max_len = x.size(0), x.size(1) - mask = seq_len_to_mask(seq_len) - mask = mask.view(batch_size, max_len) - mask = mask.to(x).float() - return mask - - def _decode(self, x): + def _decode(self, x, mask): """ :param torch.FloatTensor x: [batch_size, max_len, tag_size] :return prediction: [batch_size, max_len] """ - tag_seq, _ = self.Crf.viterbi_decode(x, self.mask) + tag_seq, _ = self.crf.viterbi_decode(x, mask) return tag_seq class AdvSeqLabel(nn.Module): """ - 别名：:class:`fastNLP.models.AdvSeqLabel` :class:`fastNLP.models.sequence_labeling.AdvSeqLabel` - 更复杂的Sequence Labelling模型。结构为Embedding, LayerNorm, 双向LSTM(两层)，FC，LayerNorm，DropOut，FC，CRF。 - :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray init_embed: Embedding的大小(传入tuple(int, int), + :param tuple(int,int),torch.FloatTensor,nn.Embedding,numpy.ndarray embed: Embedding的大小(传入tuple(int, int), 第一个int为vocab_zie, 第二个int为embed_dim); 如果为Tensor, Embedding, ndarray等则直接使用该值初始化Embedding :param int hidden_size: LSTM的隐层大小 :param int num_classes: 有多少个类 @@ -188,11 +167,11 @@ class AdvSeqLabel(nn.Module): :param str encoding_type: 支持"BIO", "BMES", "BEMSO", 只有在id2words不为None的情况有用。 """ - def __init__(self, init_embed, hidden_size, num_classes, dropout=0.3, id2words=None, encoding_type='bmes'): + def __init__(self, embed, hidden_size, num_classes, dropout=0.3, id2words=None, encoding_type='bmes'): super().__init__() - self.Embedding = embedding.Embedding(init_embed) + self.Embedding = get_embeddings(embed) self.norm1 = torch.nn.LayerNorm(self.Embedding.embedding_dim) self.Rnn = encoder.LSTM(input_size=self.Embedding.embedding_dim, hidden_size=hidden_size, num_layers=2, dropout=dropout, @@ -210,36 +189,29 @@ class AdvSeqLabel(nn.Module): allowed_transitions=allowed_transitions(id2words, encoding_type=encoding_type)) - def _decode(self, x): + def _decode(self, x, mask): """ :param torch.FloatTensor x: [batch_size, max_len, tag_size] + :param torch.ByteTensor mask: [batch_size, max_len] :return torch.LongTensor, [batch_size, max_len] """ - tag_seq, _ = self.Crf.viterbi_decode(x, self.mask) + tag_seq, _ = self.Crf.viterbi_decode(x, mask) return tag_seq - def _internal_loss(self, x, y): + def _internal_loss(self, x, y, mask): """ Negative log likelihood loss. :param x: Tensor, [batch_size, max_len, tag_size] :param y: Tensor, [batch_size, max_len] + :param mask: Tensor, [batch_size, max_len] :return loss: a scalar Tensor """ x = x.float() y = y.long() - assert x.shape[:2] == y.shape - assert y.shape == self.mask.shape - total_loss = self.Crf(x, y, self.mask) + total_loss = self.Crf(x, y, mask) return torch.mean(total_loss) - def _make_mask(self, x, seq_len): - batch_size, max_len = x.size(0), x.size(1) - mask = seq_len_to_mask(seq_len) - mask = mask.view(batch_size, max_len) - mask = mask.to(x).float() - return mask - def _forward(self, words, seq_len, target=None): """ :param torch.LongTensor words: [batch_size, mex_len] @@ -251,15 +223,13 @@ class AdvSeqLabel(nn.Module): words = words.long() seq_len = seq_len.long() - self.mask = self._make_mask(words, seq_len) - - # seq_len = seq_len.long() + mask = seq_len_to_mask(seq_len, max_len=words.size(1)) + target = target.long() if target is not None else None if next(self.parameters()).is_cuda: words = words.cuda() - self.mask = self.mask.cuda() - + x = self.Embedding(words) x = self.norm1(x) # [batch_size, max_len, word_emb_dim] @@ -272,9 +242,9 @@ class AdvSeqLabel(nn.Module): x = self.drop(x) x = self.Linear2(x) if target is not None: - return {"loss": self._internal_loss(x, target)} + return {"loss": self._internal_loss(x, target, mask)} else: - return {"pred": self._decode(x)} + return {"pred": self._decode(x, mask)} def forward(self, words, seq_len, target): """ diff --git a/fastNLP/models/snli.py b/fastNLP/models/snli.py index 8e35b6bc..07303ddc 100644 --- a/fastNLP/models/snli.py +++ b/fastNLP/models/snli.py @@ -1,3 +1,7 @@ +""" +.. todo:: + doc +""" __all__ = [ "ESIM" ] @@ -5,34 +9,34 @@ __all__ = [ import torch import torch.nn as nn import torch.nn.functional as F - from torch.nn import CrossEntropyLoss from .base_model import BaseModel -from ..embeddings.embedding import TokenEmbedding from ..core.const import Const from ..core.utils import seq_len_to_mask +from ..embeddings.embedding import TokenEmbedding, Embedding class ESIM(BaseModel): """ - 别名：:class:`fastNLP.models.ESIM` :class:`fastNLP.models.snli.ESIM` - ESIM model的一个PyTorch实现 论文参见： https://arxiv.org/pdf/1609.06038.pdf - :param fastNLP.TokenEmbedding init_embedding: 初始化的TokenEmbedding + :param embed: 初始化的Embedding :param int hidden_size: 隐藏层大小，默认值为Embedding的维度 :param int num_labels: 目标标签种类数量，默认值为3 :param float dropout_rate: dropout的比率，默认值为0.3 :param float dropout_embed: 对Embedding的dropout比率，默认值为0.1 """ - def __init__(self, init_embedding: TokenEmbedding, hidden_size=None, num_labels=3, dropout_rate=0.3, + def __init__(self, embed, hidden_size=None, num_labels=3, dropout_rate=0.3, dropout_embed=0.1): super(ESIM, self).__init__() - self.embedding = init_embedding + if isinstance(embed, TokenEmbedding) or isinstance(embed, Embedding): + self.embedding = embed + else: + self.embedding = Embedding(embed) self.dropout_embed = EmbedDropout(p=dropout_embed) if hidden_size is None: hidden_size = self.embedding.embed_size diff --git a/fastNLP/models/star_transformer.py b/fastNLP/models/star_transformer.py index b95d1c25..e4d5af84 100644 --- a/fastNLP/models/star_transformer.py +++ b/fastNLP/models/star_transformer.py @@ -19,11 +19,9 @@ from ..core.const import Const class StarTransEnc(nn.Module): """ - 别名：:class:`fastNLP.models.StarTransEnc` :class:`fastNLP.models.star_transformer.StarTransEnc` - 带word embedding的Star-Transformer Encoder - :param init_embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 + :param embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 embedding的大小和每个词的维度. 也可以传入 nn.Embedding 对象, 此时就以传入的对象作为embedding :param hidden_size: 模型中特征维度. @@ -35,7 +33,7 @@ class StarTransEnc(nn.Module): :param dropout: 模型除词嵌入外的dropout概率. """ - def __init__(self, init_embed, + def __init__(self, embed, hidden_size, num_layers, num_head, @@ -44,7 +42,7 @@ class StarTransEnc(nn.Module): emb_dropout, dropout): super(StarTransEnc, self).__init__() - self.embedding = get_embeddings(init_embed) + self.embedding = get_embeddings(embed) emb_dim = self.embedding.embedding_dim self.emb_fc = nn.Linear(emb_dim, hidden_size) # self.emb_drop = nn.Dropout(emb_dropout) @@ -104,11 +102,9 @@ class _NLICls(nn.Module): class STSeqLabel(nn.Module): """ - 别名：:class:`fastNLP.models.STSeqLabel` :class:`fastNLP.models.star_transformer.STSeqLabel` - 用于序列标注的Star-Transformer模型 - :param init_embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 + :param embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 embedding的大小和每个词的维度. 也可以传入 nn.Embedding 对象, 此时就以传入的对象作为embedding :param num_cls: 输出类别个数 @@ -122,7 +118,7 @@ class STSeqLabel(nn.Module): :param dropout: 模型除词嵌入外的dropout概率. Default: 0.1 """ - def __init__(self, init_embed, num_cls, + def __init__(self, embed, num_cls, hidden_size=300, num_layers=4, num_head=8, @@ -132,7 +128,7 @@ class STSeqLabel(nn.Module): emb_dropout=0.1, dropout=0.1, ): super(STSeqLabel, self).__init__() - self.enc = StarTransEnc(init_embed=init_embed, + self.enc = StarTransEnc(embed=embed, hidden_size=hidden_size, num_layers=num_layers, num_head=num_head, @@ -169,11 +165,9 @@ class STSeqLabel(nn.Module): class STSeqCls(nn.Module): """ - 别名：:class:`fastNLP.models.STSeqCls` :class:`fastNLP.models.star_transformer.STSeqCls` - 用于分类任务的Star-Transformer - :param init_embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 + :param embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 embedding的大小和每个词的维度. 也可以传入 nn.Embedding 对象, 此时就以传入的对象作为embedding :param num_cls: 输出类别个数 @@ -187,7 +181,7 @@ class STSeqCls(nn.Module): :param dropout: 模型除词嵌入外的dropout概率. Default: 0.1 """ - def __init__(self, init_embed, num_cls, + def __init__(self, embed, num_cls, hidden_size=300, num_layers=4, num_head=8, @@ -197,7 +191,7 @@ class STSeqCls(nn.Module): emb_dropout=0.1, dropout=0.1, ): super(STSeqCls, self).__init__() - self.enc = StarTransEnc(init_embed=init_embed, + self.enc = StarTransEnc(embed=embed, hidden_size=hidden_size, num_layers=num_layers, num_head=num_head, @@ -234,11 +228,9 @@ class STSeqCls(nn.Module): class STNLICls(nn.Module): """ - 别名：:class:`fastNLP.models.STNLICls` :class:`fastNLP.models.star_transformer.STNLICls` - 用于自然语言推断(NLI)的Star-Transformer - :param init_embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 + :param embed: 单词词典, 可以是 tuple, 包括(num_embedings, embedding_dim), 即 embedding的大小和每个词的维度. 也可以传入 nn.Embedding 对象, 此时就以传入的对象作为embedding :param num_cls: 输出类别个数 @@ -252,7 +244,7 @@ class STNLICls(nn.Module): :param dropout: 模型除词嵌入外的dropout概率. Default: 0.1 """ - def __init__(self, init_embed, num_cls, + def __init__(self, embed, num_cls, hidden_size=300, num_layers=4, num_head=8, @@ -262,7 +254,7 @@ class STNLICls(nn.Module): emb_dropout=0.1, dropout=0.1, ): super(STNLICls, self).__init__() - self.enc = StarTransEnc(init_embed=init_embed, + self.enc = StarTransEnc(embed=embed, hidden_size=hidden_size, num_layers=num_layers, num_head=num_head, diff --git a/fastNLP/modules/__init__.py b/fastNLP/modules/__init__.py index 7959e454..769dc42a 100644 --- a/fastNLP/modules/__init__.py +++ b/fastNLP/modules/__init__.py @@ -54,3 +54,7 @@ from . import encoder from .decoder import * from .dropout import TimestepDropout from .encoder import * + +import sys +from ..doc_utils import doc_process +doc_process(sys.modules[__name__]) diff --git a/fastNLP/modules/decoder/__init__.py b/fastNLP/modules/decoder/__init__.py index 664618b2..57acb172 100644 --- a/fastNLP/modules/decoder/__init__.py +++ b/fastNLP/modules/decoder/__init__.py @@ -1,3 +1,7 @@ +""" +.. todo:: + doc +""" __all__ = [ "MLP", "ConditionalRandomField", @@ -6,6 +10,6 @@ __all__ = [ ] from .crf import ConditionalRandomField +from .crf import allowed_transitions from .mlp import MLP from .utils import viterbi_decode -from .crf import allowed_transitions diff --git a/fastNLP/modules/decoder/crf.py b/fastNLP/modules/decoder/crf.py index 7c496868..aeb73d76 100644 --- a/fastNLP/modules/decoder/crf.py +++ b/fastNLP/modules/decoder/crf.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "ConditionalRandomField", "allowed_transitions" @@ -7,31 +9,44 @@ import torch from torch import nn from ..utils import initial_parameter +from ...core.vocabulary import Vocabulary +from ...core.metrics import _get_encoding_type_from_tag_vocab, _check_tag_vocab_and_encoding_type +from typing import Union - -def allowed_transitions(id2target, encoding_type='bio', include_start_end=False): +def allowed_transitions(tag_vocab:Union[Vocabulary, dict], encoding_type=None, include_start_end=False): """ - 别名：:class:`fastNLP.modules.allowed_transitions` :class:`fastNLP.modules.decoder.allowed_transitions` - 给定一个id到label的映射表，返回所有可以跳转的(from_tag_id, to_tag_id)列表。 - :param dict id2target: key是label的indices，value是str类型的tag或tag-label。value可以是只有tag的, 比如"B", "M"; 也可以是 - "B-NN", "M-NN", tag和label之间一定要用"-"隔开。一般可以通过Vocabulary.idx2word得到id2label。 - :param str encoding_type: 支持"bio", "bmes", "bmeso", "bioes"。 + :param ~fastNLP.Vocabulary,dict tag_vocab: 支持类型为tag或tag-label。只有tag的,比如"B", "M"; 也可以是"B-NN", "M-NN", + tag和label之间一定要用"-"隔开。如果传入dict，格式需要形如{0:"O", 1:"B-tag1"}，即index在前，tag在后。 + :param str encoding_type: 支持"bio", "bmes", "bmeso", "bioes"。默认为None，通过vocab自动推断 :param bool include_start_end: 是否包含开始与结尾的转换。比如在bio中，b/o可以在开头，但是i不能在开头； 为True，返回的结果中会包含(start_idx, b_idx), (start_idx, o_idx), 但是不包含(start_idx, i_idx); start_idx=len(id2label), end_idx=len(id2label)+1。为False, 返回的结果中不含与开始结尾相关的内容 :return: List[Tuple(int, int)]], 内部的Tuple是可以进行跳转的(from_tag_id, to_tag_id)。 """ - num_tags = len(id2target) + if encoding_type is None: + encoding_type = _get_encoding_type_from_tag_vocab(tag_vocab) + else: + encoding_type = encoding_type.lower() + _check_tag_vocab_and_encoding_type(tag_vocab, encoding_type) + + pad_token = '' + unk_token = '' + + if isinstance(tag_vocab, Vocabulary): + id_label_lst = list(tag_vocab.idx2word.items()) + pad_token = tag_vocab.padding + unk_token = tag_vocab.unknown + else: + id_label_lst = list(tag_vocab.items()) + + num_tags = len(tag_vocab) start_idx = num_tags end_idx = num_tags + 1 - encoding_type = encoding_type.lower() allowed_trans = [] - id_label_lst = list(id2target.items()) if include_start_end: id_label_lst += [(start_idx, 'start'), (end_idx, 'end')] - def split_tag_label(from_label): from_label = from_label.lower() if from_label in ['start', 'end']: @@ -43,11 +58,11 @@ def allowed_transitions(id2target, encoding_type='bio', include_start_end=False) return from_tag, from_label for from_id, from_label in id_label_lst: - if from_label in ['', '']: + if from_label in [pad_token, unk_token]: continue from_tag, from_label = split_tag_label(from_label) for to_id, to_label in id_label_lst: - if to_label in ['', '']: + if to_label in [pad_token, unk_token]: continue to_tag, to_label = split_tag_label(to_label) if _is_transition_allowed(encoding_type, from_tag, from_label, to_tag, to_label): @@ -151,10 +166,7 @@ def _is_transition_allowed(encoding_type, from_tag, from_label, to_tag, to_label class ConditionalRandomField(nn.Module): """ - 别名：:class:`fastNLP.modules.ConditionalRandomField` :class:`fastNLP.modules.decoder.ConditionalRandomField` - - 条件随机场。 - 提供forward()以及viterbi_decode()两个方法，分别用于训练与inference。 + 条件随机场。提供forward()以及viterbi_decode()两个方法，分别用于训练与inference。 :param int num_tags: 标签的数量 :param bool include_start_end_trans: 是否考虑各个tag作为开始以及结尾的分数。 @@ -208,7 +220,7 @@ class ConditionalRandomField(nn.Module): trans_score = self.trans_m.view(1, n_tags, n_tags) tmp = alpha.view(batch_size, n_tags, 1) + emit_score + trans_score alpha = torch.logsumexp(tmp, 1).masked_fill(flip_mask[i].view(batch_size, 1), 0) + \ - alpha.masked_fill(mask[i].byte().view(batch_size, 1), 0) + alpha.masked_fill(mask[i].eq(1).view(batch_size, 1), 0) if self.include_start_end_trans: alpha = alpha + self.end_scores.view(1, -1) @@ -228,7 +240,7 @@ class ConditionalRandomField(nn.Module): seq_idx = torch.arange(seq_len, dtype=torch.long, device=logits.device) # trans_socre [L-1, B] - mask = mask.byte() + mask = mask.eq(1) flip_mask = mask.eq(0) trans_score = self.trans_m[tags[:seq_len - 1], tags[1:]].masked_fill(flip_mask[1:, :], 0) # emit_score [L, B] @@ -276,7 +288,7 @@ class ConditionalRandomField(nn.Module): """ batch_size, seq_len, n_tags = logits.size() logits = logits.transpose(0, 1).data # L, B, H - mask = mask.transpose(0, 1).data.byte() # L, B + mask = mask.transpose(0, 1).data.eq(1) # L, B # dp vpath = logits.new_zeros((seq_len, batch_size, n_tags), dtype=torch.long) diff --git a/fastNLP/modules/decoder/mlp.py b/fastNLP/modules/decoder/mlp.py index 9d9d80f2..3e594de1 100644 --- a/fastNLP/modules/decoder/mlp.py +++ b/fastNLP/modules/decoder/mlp.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "MLP" ] @@ -10,8 +12,6 @@ from ..utils import initial_parameter class MLP(nn.Module): """ - 别名：:class:`fastNLP.modules.MLP` :class:`fastNLP.modules.decoder.MLP` - 多层感知器 :param List[int] size_layer: 一个int的列表，用来定义MLP的层数，列表中的数字为每一层是hidden数目。MLP的层数为 len(size_layer) - 1 diff --git a/fastNLP/modules/decoder/utils.py b/fastNLP/modules/decoder/utils.py index 9e773336..e0d2af68 100644 --- a/fastNLP/modules/decoder/utils.py +++ b/fastNLP/modules/decoder/utils.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "viterbi_decode" ] @@ -6,8 +8,6 @@ import torch def viterbi_decode(logits, transitions, mask=None, unpad=False): r""" - 别名：:class:`fastNLP.modules.viterbi_decode` :class:`fastNLP.modules.decoder.viterbi_decode` - 给定一个特征矩阵以及转移分数矩阵，计算出最佳的路径以及对应的分数 :param torch.FloatTensor logits: batch_size x max_len x num_tags，特征矩阵。 @@ -27,7 +27,7 @@ def viterbi_decode(logits, transitions, mask=None, unpad=False): "compatible." logits = logits.transpose(0, 1).data # L, B, H if mask is not None: - mask = mask.transpose(0, 1).data.byte() # L, B + mask = mask.transpose(0, 1).data.eq(1) # L, B else: mask = logits.new_ones((seq_len, batch_size), dtype=torch.uint8) diff --git a/fastNLP/modules/dropout.py b/fastNLP/modules/dropout.py index 0ea2a2d9..24c20cc6 100644 --- a/fastNLP/modules/dropout.py +++ b/fastNLP/modules/dropout.py @@ -1,4 +1,8 @@ -__all__ = [] +"""undocumented""" + +__all__ = [ + "TimestepDropout" +] import torch diff --git a/fastNLP/modules/encoder/__init__.py b/fastNLP/modules/encoder/__init__.py index 1e99a0fd..0dfc18de 100644 --- a/fastNLP/modules/encoder/__init__.py +++ b/fastNLP/modules/encoder/__init__.py @@ -1,3 +1,8 @@ +""" +.. todo:: + doc +""" + __all__ = [ # "BertModel", @@ -24,13 +29,12 @@ __all__ = [ "MultiHeadAttention", ] +from .attention import MultiHeadAttention from .bert import BertModel from .char_encoder import ConvolutionCharEncoder, LSTMCharEncoder from .conv_maxpool import ConvMaxpool from .lstm import LSTM +from .pooling import MaxPool, MaxPoolWithMask, AvgPool, AvgPoolWithMask from .star_transformer import StarTransformer from .transformer import TransformerEncoder from .variational_rnn import VarRNN, VarLSTM, VarGRU - -from .pooling import MaxPool, MaxPoolWithMask, AvgPool, AvgPoolWithMask -from .attention import MultiHeadAttention diff --git a/fastNLP/modules/encoder/_elmo.py b/fastNLP/modules/encoder/_elmo.py index befae8bc..554cf8a9 100644 --- a/fastNLP/modules/encoder/_elmo.py +++ b/fastNLP/modules/encoder/_elmo.py @@ -1,7 +1,9 @@ -""" +"""undocumented 这个页面的代码大量参考了 allenNLP """ +__all__ = [] + from typing import Optional, Tuple, List, Callable import torch diff --git a/fastNLP/modules/encoder/attention.py b/fastNLP/modules/encoder/attention.py index fe3f7fd8..0d832653 100644 --- a/fastNLP/modules/encoder/attention.py +++ b/fastNLP/modules/encoder/attention.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "MultiHeadAttention" ] @@ -28,14 +30,14 @@ class DotAttention(nn.Module): def forward(self, Q, K, V, mask_out=None): """ - :param Q: [batch, seq_len_q, key_size] - :param K: [batch, seq_len_k, key_size] - :param V: [batch, seq_len_k, value_size] - :param mask_out: [batch, 1, seq_len] or [batch, seq_len_q, seq_len_k] + :param Q: [..., seq_len_q, key_size] + :param K: [..., seq_len_k, key_size] + :param V: [..., seq_len_k, value_size] + :param mask_out: [..., 1, seq_len] or [..., seq_len_q, seq_len_k] """ - output = torch.matmul(Q, K.transpose(1, 2)) / self.scale + output = torch.matmul(Q, K.transpose(-1, -2)) / self.scale if mask_out is not None: - output.masked_fill_(mask_out, -1e18) + output.masked_fill_(mask_out, -1e9) output = self.softmax(output) output = self.drop(output) return torch.matmul(output, V) @@ -43,7 +45,6 @@ class DotAttention(nn.Module): class MultiHeadAttention(nn.Module): """ - 别名：:class:`fastNLP.modules.MultiHeadAttention` :class:`fastNLP.modules.encoder.MultiHeadAttention` :param input_size: int, 输入维度的大小。同时也是输出维度的大小。 :param key_size: int, 每个head的维度大小。 @@ -63,17 +64,16 @@ class MultiHeadAttention(nn.Module): self.q_in = nn.Linear(input_size, in_size) self.k_in = nn.Linear(input_size, in_size) self.v_in = nn.Linear(input_size, in_size) - # follow the paper, do not apply dropout within dot-product self.attention = DotAttention(key_size=key_size, value_size=value_size, dropout=dropout) self.out = nn.Linear(value_size * num_head, input_size) self.reset_parameters() def reset_parameters(self): sqrt = math.sqrt - nn.init.normal_(self.q_in.weight, mean=0, std=sqrt(2.0 / (self.input_size + self.key_size))) - nn.init.normal_(self.k_in.weight, mean=0, std=sqrt(2.0 / (self.input_size + self.key_size))) - nn.init.normal_(self.v_in.weight, mean=0, std=sqrt(2.0 / (self.input_size + self.value_size))) - nn.init.xavier_normal_(self.out.weight) + nn.init.normal_(self.q_in.weight, mean=0, std=sqrt(1.0 / self.input_size)) + nn.init.normal_(self.k_in.weight, mean=0, std=sqrt(1.0 / self.input_size)) + nn.init.normal_(self.v_in.weight, mean=0, std=sqrt(1.0 / self.input_size)) + nn.init.normal_(self.out.weight, mean=0, std=sqrt(1.0 / self.input_size)) def forward(self, Q, K, V, atte_mask_out=None): """ @@ -87,20 +87,16 @@ class MultiHeadAttention(nn.Module): sk = K.size(1) d_k, d_v, n_head = self.key_size, self.value_size, self.num_head # input linear - q = self.q_in(Q).view(batch, sq, n_head, d_k) - k = self.k_in(K).view(batch, sk, n_head, d_k) - v = self.v_in(V).view(batch, sk, n_head, d_v) - - # transpose q, k and v to do batch attention - q = q.permute(2, 0, 1, 3).contiguous().view(-1, sq, d_k) - k = k.permute(2, 0, 1, 3).contiguous().view(-1, sk, d_k) - v = v.permute(2, 0, 1, 3).contiguous().view(-1, sk, d_v) + q = self.q_in(Q).view(batch, sq, n_head, d_k).transpose(1, 2) + k = self.k_in(K).view(batch, sk, n_head, d_k).transpose(1, 2) + v = self.v_in(V).view(batch, sk, n_head, d_v).transpose(1, 2) + if atte_mask_out is not None: - atte_mask_out = atte_mask_out.repeat(n_head, 1, 1) - atte = self.attention(q, k, v, atte_mask_out).view(n_head, batch, sq, d_v) + atte_mask_out = atte_mask_out[:,None,:,:] # [bsz,1,1,len] + atte = self.attention(q, k, v, atte_mask_out).view(batch, n_head, sq, d_v) # concat all heads, do output linear - atte = atte.permute(1, 2, 0, 3).contiguous().view(batch, sq, -1) + atte = atte.transpose(1, 2).contiguous().view(batch, sq, -1) output = self.out(atte) return output diff --git a/fastNLP/modules/encoder/bert.py b/fastNLP/modules/encoder/bert.py index ce175df1..12379718 100644 --- a/fastNLP/modules/encoder/bert.py +++ b/fastNLP/modules/encoder/bert.py @@ -1,4 +1,4 @@ -""" +"""undocumented 这个页面的代码很大程度上参考(复制粘贴)了https://github.com/huggingface/pytorch-pretrained-BERT的代码， 如果你发现该代码对你 有用，也请引用一下他们。 """ @@ -8,24 +8,23 @@ __all__ = [ ] import collections - -import unicodedata import copy import json import math import os +import unicodedata import torch from torch import nn -import sys from ..utils import _get_file_name_base_on_postfix +from ...io.file_utils import _get_embedding_url, cached_path, PRETRAINED_BERT_MODEL_DIR +from ...core import logger CONFIG_FILE = 'bert_config.json' VOCAB_NAME = 'vocab.txt' - class BertConfig(object): """Configuration class to store the configuration of a `BertModel`. """ @@ -134,6 +133,19 @@ def swish(x): ACT2FN = {"gelu": gelu, "relu": torch.nn.functional.relu, "swish": swish} +def _get_bert_dir(model_dir_or_name: str = 'en-base-uncased'): + if model_dir_or_name.lower() in PRETRAINED_BERT_MODEL_DIR: + model_url = _get_embedding_url('bert', model_dir_or_name.lower()) + model_dir = cached_path(model_url, name='embedding') + # 检查是否存在 + elif os.path.isdir(os.path.abspath(os.path.expanduser(model_dir_or_name))): + model_dir = os.path.abspath(os.path.expanduser(model_dir_or_name)) + else: + logger.error(f"Cannot recognize BERT dir or name ``{model_dir_or_name}``.") + raise ValueError(f"Cannot recognize BERT dir or name ``{model_dir_or_name}``.") + return str(model_dir) + + class BertLayerNorm(nn.Module): def __init__(self, hidden_size, eps=1e-12): """Construct a layernorm module in the TF style (epsilon inside the square root). @@ -336,31 +348,11 @@ class BertPooler(nn.Module): class BertModel(nn.Module): """ - 别名：:class:`fastNLP.modules.BertModel` :class:`fastNLP.modules.encoder.BertModel` - BERT(Bidirectional Embedding Representations from Transformers). - 如果你想使用预训练好的权重矩阵，请在以下网址下载. - sources:: - - 'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-pytorch_model.bin", - 'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-pytorch_model.bin", - 'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-pytorch_model.bin", - 'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-pytorch_model.bin", - 'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased-pytorch_model.bin", - 'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-pytorch_model.bin", - 'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-pytorch_model.bin", - 'bert-base-german-cased': "https://int-deepset-models-bert.s3.eu-central-1.amazonaws.com/pytorch/bert-base-german-cased-pytorch_model.bin", - 'bert-large-uncased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-pytorch_model.bin", - 'bert-large-cased-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-pytorch_model.bin", - 'bert-large-uncased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-whole-word-masking-finetuned-squad-pytorch_model.bin", - 'bert-large-cased-whole-word-masking-finetuned-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-whole-word-masking-finetuned-squad-pytorch_model.bin", - 'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-pytorch_model.bin" - - 用预训练权重矩阵来建立BERT模型:: - model = BertModel.from_pretrained("path/to/weights/directory") + model = BertModel.from_pretrained(model_dir_or_name) 用随机初始化权重矩阵来建立BERT模型:: @@ -441,11 +433,15 @@ class BertModel(nn.Module): return encoded_layers, pooled_output @classmethod - def from_pretrained(cls, pretrained_model_dir, *inputs, **kwargs): + def from_pretrained(cls, model_dir_or_name, *inputs, **kwargs): state_dict = kwargs.get('state_dict', None) kwargs.pop('state_dict', None) kwargs.pop('cache_dir', None) kwargs.pop('from_tf', None) + + # get model dir from name or dir + pretrained_model_dir = _get_bert_dir(model_dir_or_name) + # Load config config_file = _get_file_name_base_on_postfix(pretrained_model_dir, '.json') config = BertConfig.from_json_file(config_file) @@ -455,6 +451,9 @@ class BertModel(nn.Module): if state_dict is None: weights_path = _get_file_name_base_on_postfix(pretrained_model_dir, '.bin') state_dict = torch.load(weights_path, map_location='cpu') + else: + logger.error(f'Cannot load parameters through `state_dict` variable.') + raise RuntimeError(f'Cannot load parameters through `state_dict` variable.') old_keys = [] new_keys = [] @@ -489,11 +488,13 @@ class BertModel(nn.Module): load(model, prefix='' if hasattr(model, 'bert') else 'bert.') if len(missing_keys) > 0: - print("Weights of {} not initialized from pretrained model: {}".format( + logger.warn("Weights of {} not initialized from pretrained model: {}".format( model.__class__.__name__, missing_keys)) if len(unexpected_keys) > 0: - print("Weights from pretrained model not used in {}: {}".format( + logger.warn("Weights from pretrained model not used in {}: {}".format( model.__class__.__name__, unexpected_keys)) + + logger.info(f"Load pre-trained BERT parameters from file {weights_path}.") return model @@ -563,6 +564,8 @@ class WordpieceTokenizer(object): output_tokens.append(self.unk_token) else: output_tokens.extend(sub_tokens) + if len(output_tokens) == 0: # 防止里面全是空格或者回车符号 + return [self.unk_token] return output_tokens @@ -672,14 +675,14 @@ class BasicTokenizer(object): # as is Japanese Hiragana and Katakana. Those alphabets are used to write # space-separated words, so they are not treated specially and handled # like the all of the other languages. - if ((cp >= 0x4E00 and cp <= 0x9FFF) or # - (cp >= 0x3400 and cp <= 0x4DBF) or # - (cp >= 0x20000 and cp <= 0x2A6DF) or # - (cp >= 0x2A700 and cp <= 0x2B73F) or # - (cp >= 0x2B740 and cp <= 0x2B81F) or # - (cp >= 0x2B820 and cp <= 0x2CEAF) or - (cp >= 0xF900 and cp <= 0xFAFF) or # - (cp >= 0x2F800 and cp <= 0x2FA1F)): # + if (((cp >= 0x4E00) and (cp <= 0x9FFF)) or # + ((cp >= 0x3400) and (cp <= 0x4DBF)) or # + ((cp >= 0x20000) and (cp <= 0x2A6DF)) or # + ((cp >= 0x2A700) and (cp <= 0x2B73F)) or # + ((cp >= 0x2B740) and (cp <= 0x2B81F)) or # + ((cp >= 0x2B820) and (cp <= 0x2CEAF)) or + ((cp >= 0xF900) and (cp <= 0xFAFF)) or # + ((cp >= 0x2F800) and (cp <= 0x2FA1F))): # return True return False @@ -729,8 +732,8 @@ def _is_punctuation(char): # Characters such as "^", "$", and "`" are not in the Unicode # Punctuation class but we treat them as punctuation anyways, for # consistency. - if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or - (cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)): + if (((cp >= 33) and (cp <= 47)) or ((cp >= 58) and (cp <= 64)) or + ((cp >= 91) and (cp <= 96)) or ((cp >= 123) and (cp <= 126))): return True cat = unicodedata.category(char) if cat.startswith("P"): @@ -797,7 +800,7 @@ class BertTokenizer(object): for token in tokens: ids.append(self.vocab[token]) if len(ids) > self.max_len: - print( + logger.warn( "Token indices sequence length is longer than the specified maximum " " sequence length for this BERT model ({} > {}). Running this" " sequence through BERT will result in indexing errors".format(len(ids), self.max_len) @@ -821,7 +824,7 @@ class BertTokenizer(object): with open(vocab_file, "w", encoding="utf-8") as writer: for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]): if index != token_index: - print("Saving vocabulary to {}: vocabulary indices are not consecutive." + logger.warn("Saving vocabulary to {}: vocabulary indices are not consecutive." " Please check that the vocabulary is not corrupted!".format(vocab_file)) index = token_index writer.write(token + u'\n') @@ -829,30 +832,31 @@ class BertTokenizer(object): return vocab_file @classmethod - def from_pretrained(cls, model_dir, *inputs, **kwargs): + def from_pretrained(cls, model_dir_or_name, *inputs, **kwargs): """ - 给定path，直接读取vocab. - + 给定模型的名字或者路径，直接读取vocab. """ + model_dir = _get_bert_dir(model_dir_or_name) pretrained_model_name_or_path = _get_file_name_base_on_postfix(model_dir, '.txt') - print("loading vocabulary file {}".format(pretrained_model_name_or_path)) + logger.info("loading vocabulary file {}".format(pretrained_model_name_or_path)) max_len = 512 kwargs['max_len'] = min(kwargs.get('max_position_embeddings', int(1e12)), max_len) # Instantiate tokenizer. tokenizer = cls(pretrained_model_name_or_path, *inputs, **kwargs) return tokenizer + class _WordPieceBertModel(nn.Module): """ 这个模块用于直接计算word_piece的结果. """ - def __init__(self, model_dir: str, layers: str = '-1'): + def __init__(self, model_dir_or_name: str, layers: str = '-1', pooled_cls: bool=False): super().__init__() - self.tokenzier = BertTokenizer.from_pretrained(model_dir) - self.encoder = BertModel.from_pretrained(model_dir) + self.tokenzier = BertTokenizer.from_pretrained(model_dir_or_name) + self.encoder = BertModel.from_pretrained(model_dir_or_name) # 检查encoder_layer_number是否合理 encoder_layer_number = len(self.encoder.encoder.layer) self.layers = list(map(int, layers.split(','))) @@ -866,9 +870,11 @@ class _WordPieceBertModel(nn.Module): self._cls_index = self.tokenzier.vocab['[CLS]'] self._sep_index = self.tokenzier.vocab['[SEP]'] + self._wordpiece_unknown_index = self.tokenzier.vocab['[UNK]'] self._wordpiece_pad_index = self.tokenzier.vocab['[PAD]'] # 需要用于生成word_piece + self.pooled_cls = pooled_cls - def index_dataset(self, *datasets, field_name): + def index_dataset(self, *datasets, field_name, add_cls_sep=True): """ 使用bert的tokenizer新生成word_pieces列加入到datasets中，并将他们设置为input。如果首尾不是 [CLS]与[SEP]会在首尾额外加入[CLS]与[SEP], 且将word_pieces这一列的pad value设置为了bert的pad value。 @@ -884,10 +890,11 @@ class _WordPieceBertModel(nn.Module): tokens = self.tokenzier.wordpiece_tokenizer.tokenize(word) word_piece_ids = self.tokenzier.convert_tokens_to_ids(tokens) word_pieces.extend(word_piece_ids) - if word_pieces[0] != self._cls_index: - word_pieces.insert(0, self._cls_index) - if word_pieces[-1] != self._sep_index: - word_pieces.insert(-1, self._sep_index) + if add_cls_sep: + if word_pieces[0] != self._cls_index: + word_pieces.insert(0, self._cls_index) + if word_pieces[-1] != self._sep_index: + word_pieces.insert(-1, self._sep_index) return word_pieces for index, dataset in enumerate(datasets): @@ -896,7 +903,7 @@ class _WordPieceBertModel(nn.Module): is_input=True) dataset.set_pad_val('word_pieces', self._wordpiece_pad_index) except Exception as e: - print(f"Exception happens when processing the {index} dataset.") + logger.error(f"Exception happens when processing the {index} dataset.") raise e def forward(self, word_pieces, token_type_ids=None): @@ -909,10 +916,13 @@ class _WordPieceBertModel(nn.Module): batch_size, max_len = word_pieces.size() attn_masks = word_pieces.ne(self._wordpiece_pad_index) - bert_outputs, _ = self.encoder(word_pieces, token_type_ids=token_type_ids, attention_mask=attn_masks, - output_all_encoded_layers=True) + bert_outputs, pooled_cls = self.encoder(word_pieces, token_type_ids=token_type_ids, attention_mask=attn_masks, + output_all_encoded_layers=True) # output_layers = [self.layers] # len(self.layers) x batch_size x max_word_piece_length x hidden_size outputs = bert_outputs[0].new_zeros((len(self.layers), batch_size, max_len, bert_outputs[0].size(-1))) for l_index, l in enumerate(self.layers): - outputs[l_index] = bert_outputs[l] + bert_output = bert_outputs[l] + if l in (len(bert_outputs)-1, -1) and self.pooled_cls: + bert_output[:, 0] = pooled_cls + outputs[l_index] = bert_output return outputs diff --git a/fastNLP/modules/encoder/char_encoder.py b/fastNLP/modules/encoder/char_encoder.py index 6a6e1470..dc73f447 100644 --- a/fastNLP/modules/encoder/char_encoder.py +++ b/fastNLP/modules/encoder/char_encoder.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "ConvolutionCharEncoder", "LSTMCharEncoder" @@ -11,8 +13,6 @@ from ..utils import initial_parameter # from torch.nn.init import xavier_uniform class ConvolutionCharEncoder(nn.Module): """ - 别名：:class:`fastNLP.modules.ConvolutionCharEncoder` :class:`fastNLP.modules.encoder.ConvolutionCharEncoder` - char级别的卷积编码器. :param int char_emb_size: char级别embedding的维度. Default: 50 @@ -58,11 +58,7 @@ class ConvolutionCharEncoder(nn.Module): class LSTMCharEncoder(nn.Module): """ - 别名：:class:`fastNLP.modules.LSTMCharEncoder` :class:`fastNLP.modules.encoder.LSTMCharEncoder` - char级别基于LSTM的encoder. - - """ def __init__(self, char_emb_size=50, hidden_size=None, initial_method=None): diff --git a/fastNLP/modules/encoder/conv_maxpool.py b/fastNLP/modules/encoder/conv_maxpool.py index 8ce6b163..bf629eba 100644 --- a/fastNLP/modules/encoder/conv_maxpool.py +++ b/fastNLP/modules/encoder/conv_maxpool.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "ConvMaxpool" ] @@ -8,8 +10,6 @@ import torch.nn.functional as F class ConvMaxpool(nn.Module): """ - 别名：:class:`fastNLP.modules.ConvMaxpool` :class:`fastNLP.modules.encoder.ConvMaxpool` - 集合了Convolution和Max-Pooling于一体的层。给定一个batch_size x max_len x input_size的输入，返回batch_size x sum(output_channels) 大小的matrix。在内部，是先使用CNN给输入做卷积，然后经过activation激活层，在通过在长度(max_len) 这一维进行max_pooling。最后得到每个sample的一个向量表示。 diff --git a/fastNLP/modules/encoder/lstm.py b/fastNLP/modules/encoder/lstm.py index e2358132..1dd1f0df 100644 --- a/fastNLP/modules/encoder/lstm.py +++ b/fastNLP/modules/encoder/lstm.py @@ -1,7 +1,8 @@ -""" +"""undocumented 轻量封装的 Pytorch LSTM 模块. 可在 forward 时传入序列的长度, 自动对padding做合适的处理. """ + __all__ = [ "LSTM" ] @@ -13,8 +14,6 @@ import torch.nn.utils.rnn as rnn class LSTM(nn.Module): """ - 别名：:class:`fastNLP.modules.LSTM` :class:`fastNLP.modules.encoder.LSTM` - LSTM 模块, 轻量封装的Pytorch LSTM. 在提供seq_len的情况下，将自动使用pack_padded_sequence; 同时默认将forget gate的bias初始化 为1; 且可以应对DataParallel中LSTM的使用问题。 diff --git a/fastNLP/modules/encoder/pooling.py b/fastNLP/modules/encoder/pooling.py index d8aa54ad..c248601d 100644 --- a/fastNLP/modules/encoder/pooling.py +++ b/fastNLP/modules/encoder/pooling.py @@ -1,3 +1,5 @@ +"""undocumented""" + __all__ = [ "MaxPool", "MaxPoolWithMask", @@ -10,8 +12,6 @@ import torch.nn as nn class MaxPool(nn.Module): """ - 别名：:class:`fastNLP.modules.MaxPool` :class:`fastNLP.modules.encoder.MaxPool` - Max-pooling模块。 :param stride: 窗口移动大小，默认为kernel_size @@ -59,8 +59,6 @@ class MaxPool(nn.Module): class MaxPoolWithMask(nn.Module): """ - 别名：:class:`fastNLP.modules.MaxPoolWithMask` :class:`fastNLP.modules.encoder.MaxPoolWithMask` - 带mask矩阵的max pooling。在做max-pooling的时候不会考虑mask值为0的位置。 """ @@ -99,8 +97,6 @@ class KMaxPool(nn.Module): class AvgPool(nn.Module): """ - 别名：:class:`fastNLP.modules.AvgPool` :class:`fastNLP.modules.encoder.AvgPool` - 给定形如[batch_size, max_len, hidden_size]的输入，在最后一维进行avg pooling. 输出为[batch_size, hidden_size] """ @@ -126,8 +122,6 @@ class AvgPool(nn.Module): class AvgPoolWithMask(nn.Module): """ - 别名：:class:`fastNLP.modules.AvgPoolWithMask` :class:`fastNLP.modules.encoder.AvgPoolWithMask` - 给定形如[batch_size, max_len, hidden_size]的输入，在最后一维进行avg pooling. 输出为[batch_size, hidden_size], pooling 的时候只会考虑mask为1的位置 """ diff --git a/fastNLP/modules/encoder/star_transformer.py b/fastNLP/modules/encoder/star_transformer.py index 3927a494..bb47d9b5 100644 --- a/fastNLP/modules/encoder/star_transformer.py +++ b/fastNLP/modules/encoder/star_transformer.py @@ -1,6 +1,7 @@ -""" +"""undocumented Star-Transformer 的encoder部分的 Pytorch 实现 """ + __all__ = [ "StarTransformer" ] @@ -13,9 +14,6 @@ from torch.nn import functional as F class StarTransformer(nn.Module): """ - 别名：:class:`fastNLP.modules.StarTransformer` :class:`fastNLP.modules.encoder.StarTransformer` - - Star-Transformer 的encoder部分。 输入3d的文本输入, 返回相同长度的文本编码 paper: https://arxiv.org/abs/1902.09113 diff --git a/fastNLP/modules/encoder/transformer.py b/fastNLP/modules/encoder/transformer.py index bc488e54..3d97c306 100644 --- a/fastNLP/modules/encoder/transformer.py +++ b/fastNLP/modules/encoder/transformer.py @@ -1,17 +1,15 @@ +"""undocumented""" + __all__ = [ "TransformerEncoder" ] from torch import nn -from fastNLP.modules.encoder.attention import MultiHeadAttention -from ..dropout import TimestepDropout +from .attention import MultiHeadAttention class TransformerEncoder(nn.Module): """ - 别名：:class:`fastNLP.modules.TransformerEncoder` :class:`fastNLP.modules.encoder.TransformerEncoder` - - transformer的encoder模块，不包含embedding层 :param int num_layers: transformer的层数 @@ -27,12 +25,13 @@ class TransformerEncoder(nn.Module): def __init__(self, model_size, inner_size, key_size, value_size, num_head, dropout=0.1): super(TransformerEncoder.SubLayer, self).__init__() self.atte = MultiHeadAttention(model_size, key_size, value_size, num_head, dropout) - self.norm1 = nn.LayerNorm(model_size) + self.norm1 = nn.LayerNorm(model_size, eps=1e-6) self.ffn = nn.Sequential(nn.Linear(model_size, inner_size), nn.ReLU(), - nn.Linear(inner_size, model_size), - TimestepDropout(dropout), ) - self.norm2 = nn.LayerNorm(model_size) + nn.Dropout(dropout), + nn.Linear(inner_size, model_size)) + self.norm2 = nn.LayerNorm(model_size, eps=1e-6) + self.dropout = nn.Dropout(dropout) def forward(self, input, seq_mask=None, atte_mask_out=None): """ @@ -41,17 +40,22 @@ class TransformerEncoder(nn.Module): :param seq_mask: [batch, seq_len] :return: [batch, seq_len, model_size] """ + if seq_mask is None: # 防止后续乘法时出错 + seq_mask = 1 + input = self.norm1(input) attention = self.atte(input, input, input, atte_mask_out) - norm_atte = self.norm1(attention + input) + input = input + self.dropout(attention) attention *= seq_mask - output = self.ffn(norm_atte) - output = self.norm2(output + norm_atte) - output *= seq_mask - return output + input = self.norm2(input) + output = self.ffn(input) + input = input + self.dropout(output) + input *= seq_mask + return input def __init__(self, num_layers, **kargs): super(TransformerEncoder, self).__init__() self.layers = nn.ModuleList([self.SubLayer(**kargs) for _ in range(num_layers)]) + self.norm = nn.LayerNorm(kargs['model_size'], eps=1e-6) def forward(self, x, seq_mask=None): """ @@ -64,8 +68,8 @@ class TransformerEncoder(nn.Module): if seq_mask is None: atte_mask_out = None else: - atte_mask_out = (seq_mask < 1)[:, None, :] + atte_mask_out = (seq_mask == 0)[:, None, :] seq_mask = seq_mask[:, :, None] for layer in self.layers: output = layer(output, seq_mask, atte_mask_out) - return output + return self.norm(output) diff --git a/fastNLP/modules/encoder/variational_rnn.py b/fastNLP/modules/encoder/variational_rnn.py index 8e5e804b..17e2ad23 100644 --- a/fastNLP/modules/encoder/variational_rnn.py +++ b/fastNLP/modules/encoder/variational_rnn.py @@ -1,6 +1,7 @@ -""" +"""undocumented Variational RNN 的 Pytorch 实现 """ + __all__ = [ "VarRNN", "VarLSTM", @@ -222,8 +223,6 @@ class VarRNNBase(nn.Module): class VarLSTM(VarRNNBase): """ - 别名：:class:`fastNLP.modules.VarLSTM` :class:`fastNLP.modules.encoder.VarLSTM` - Variational Dropout LSTM. :param input_size: 输入 `x` 的特征维度 @@ -247,8 +246,6 @@ class VarLSTM(VarRNNBase): class VarRNN(VarRNNBase): """ - 别名：:class:`fastNLP.modules.VarRNN` :class:`fastNLP.modules.encoder.VarRNN` - Variational Dropout RNN. :param input_size: 输入 `x` 的特征维度 @@ -272,8 +269,6 @@ class VarRNN(VarRNNBase): class VarGRU(VarRNNBase): """ - 别名：:class:`fastNLP.modules.VarGRU` :class:`fastNLP.modules.encoder.VarGRU` - Variational Dropout GRU. :param input_size: 输入 `x` 的特征维度 diff --git a/fastNLP/modules/utils.py b/fastNLP/modules/utils.py index dbae9c73..09574782 100644 --- a/fastNLP/modules/utils.py +++ b/fastNLP/modules/utils.py @@ -1,3 +1,14 @@ +""" +.. todo:: + doc +""" + +__all__ = [ + "initial_parameter", + "summary" +] + +import os from functools import reduce import torch @@ -39,7 +50,7 @@ def initial_parameter(net, initial_method=None): init_method = init.uniform_ else: init_method = init.xavier_normal_ - + def weights_init(m): # classname = m.__class__.__name__ if isinstance(m, nn.Conv2d) or isinstance(m, nn.Conv1d) or isinstance(m, nn.Conv3d): # for all the cnn @@ -65,7 +76,7 @@ def initial_parameter(net, initial_method=None): else: init.normal_(w.data) # bias # print("init else") - + net.apply(weights_init) @@ -78,11 +89,11 @@ def summary(model: nn.Module): """ train = [] nontrain = [] - + def layer_summary(module: nn.Module): def count_size(sizes): - return reduce(lambda x, y: x*y, sizes) - + return reduce(lambda x, y: x * y, sizes) + for p in module.parameters(recurse=False): if p.requires_grad: train.append(count_size(p.shape)) @@ -90,7 +101,7 @@ def summary(model: nn.Module): nontrain.append(count_size(p.shape)) for subm in module.children(): layer_summary(subm) - + layer_summary(model) total_train = sum(train) total_nontrain = sum(nontrain) @@ -100,7 +111,7 @@ def summary(model: nn.Module): strings.append('Trainable params: {:,}'.format(total_train)) strings.append('Non-trainable params: {:,}'.format(total_nontrain)) max_len = len(max(strings, key=len)) - bar = '-'*(max_len + 3) + bar = '-' * (max_len + 3) strings = [bar] + strings + [bar] print('\n'.join(strings)) return total, total_train, total_nontrain @@ -111,7 +122,7 @@ def get_dropout_mask(drop_p: float, tensor: torch.Tensor): 根据tensor的形状，生成一个mask :param drop_p: float, 以多大的概率置为0。 - :param tensor:torch.Tensor + :param tensor: torch.Tensor :return: torch.FloatTensor. 与tensor一样的shape """ mask_x = torch.ones_like(tensor) @@ -119,7 +130,6 @@ def get_dropout_mask(drop_p: float, tensor: torch.Tensor): training=False, inplace=True) return mask_x -import glob def _get_file_name_base_on_postfix(dir_path, postfix): """ @@ -128,9 +138,9 @@ def _get_file_name_base_on_postfix(dir_path, postfix): :param postfix: 形如".bin", ".json"等 :return: str，文件的路径 """ - files = glob.glob(os.path.join(dir_path, '*' + postfix)) + files = list(filter(lambda filename: filename.endswith(postfix), os.listdir(os.path.join(dir_path)))) if len(files) == 0: - raise FileNotFoundError(f"There is no file endswith *.{postfix} file in {dir_path}") + raise FileNotFoundError(f"There is no file endswith *{postfix} file in {dir_path}") elif len(files) > 1: - raise FileExistsError(f"There are multiple *.{postfix} files in {dir_path}") - return os.path.join(dir_path, files[0]) \ No newline at end of file + raise FileExistsError(f"There are multiple *{postfix} files in {dir_path}") + return os.path.join(dir_path, files[0]) diff --git a/legacy/api/README.md b/legacy/api/README.md deleted file mode 100644 index 73560f9f..00000000 --- a/legacy/api/README.md +++ /dev/null @@ -1,44 +0,0 @@ -# fastNLP 高级接口 - -### 环境与配置 -1. 系统环境：linux/ubuntu(推荐) -2. 编程语言：Python>=3.6 -3. Python包依赖 - - **torch==1.0** - - numpy>=1.14.2 - -### 中文分词 -```python -text = ['编者按：7月12日，英国航空航天系统公司公布了该公司研制的第一款高科技隐形无人机雷电之神。', - '这款飞行从外型上来看酷似电影中的太空飞行器，据英国方面介绍，可以实现洲际远程打击。', - '那么这款无人机到底有多厉害？'] -from fastNLP.api import CWS -cws = CWS(device='cpu') -print(cws.predict(text)) -# ['编者 按 : 7月 12日 , 英国 航空 航天 系统 公司 公布 了 该 公司 研制 的 第一 款 高 科技 隐形 无人 机雷电 之 神 。', '这 款 飞行 从 外型 上 来 看 酷似 电影 中 的 太空 飞行器 , 据 英国 方面 介绍 , 可以 实现 洲际 远程 打击 。', '那么 这 款 无人 机 到底 有 多 厉害 ?'] -``` - -### 词性标注 -```python -# 输入已分词序列 -text = [['编者', '按：', '7月', '12日', '，', '英国', '航空', '航天', '系统', '公司', '公布', '了', '该', '公司', - '研制', '的', '第一款', '高科技', '隐形', '无人机', '雷电之神', '。'], - ['那么', '这', '款', '无人机', '到底', '有', '多', '厉害', '？']] -from fastNLP.api import POS -pos = POS(device='cpu') -print(pos.predict(text)) -# [['编者/NN', '按：/NN', '7月/NT', '12日/NT', '，/PU', '英国/NR', '航空/NN', '航天/NN', '系统/NN', '公司/NN', '公布/VV', '了/AS', '该/DT', '公司/NN', '研制/VV', '的/DEC', '第一款/NN', '高科技/NN', '隐形/AD', '无人机/VV', '雷电之神/NN', '。/PU'], ['那么/AD', '这/DT', '款/NN', '无人机/VV', '到底/AD', '有/VE', '多/AD', '厉害/VA', '？/PU']] -``` - -### 句法分析 -```python -text = [['编者', '按：', '7月', '12日', '，', '英国', '航空', '航天', '系统', '公司', '公布', '了', '该', '公司', - '研制', '的', '第一款', '高科技', '隐形', '无人机', '雷电之神', '。'], - ['那么', '这', '款', '无人机', '到底', '有', '多', '厉害', '？']] -from fastNLP.api import Parser -parser = Parser(device='cpu') -print(parser.predict(text)) -# [['2/nn', '4/nn', '4/nn', '20/tmod', '11/punct', '10/nn', '10/nn', '10/nn', '10/nn', '11/nsubj', '20/dep', '11/asp', '14/det', '15/nsubj', '18/rcmod', '15/cpm', '18/nn', '11/dobj', '20/advmod', '0/root', '20/dobj', '20/punct'], ['4/advmod', '3/det', '8/xsubj', '8/dep', '8/advmod', '8/dep', '8/advmod', '0/root', '8/punct']] -``` - -完整样例见`examples.py` \ No newline at end of file diff --git a/legacy/api/__init__.py b/legacy/api/__init__.py deleted file mode 100644 index 5171d8c2..00000000 --- a/legacy/api/__init__.py +++ /dev/null @@ -1,2 +0,0 @@ -__all__ = ["CWS", "POS", "Parser"] -from .api import CWS, POS, Parser diff --git a/legacy/api/api.py b/legacy/api/api.py deleted file mode 100644 index 1408731f..00000000 --- a/legacy/api/api.py +++ /dev/null @@ -1,463 +0,0 @@ -import warnings - -import torch - -warnings.filterwarnings('ignore') -import os - -from fastNLP.core.dataset import DataSet -from .utils import load_url -from .processor import ModelProcessor -from fastNLP.io.dataset_loader import _cut_long_sentence -from fastNLP.io.data_loader import ConllLoader -from fastNLP.core.instance import Instance -from ..api.pipeline import Pipeline -from fastNLP.core.metrics import SpanFPreRecMetric -from .processor import IndexerProcessor - -# TODO add pretrain urls -model_urls = { - "cws": "http://123.206.98.91:8888/download/cws_lstm_ctb9_1_20-09908656.pkl", - "pos": "http://123.206.98.91:8888/download/pos_tag_model_20190119-43f8b435.pkl", - "parser": "http://123.206.98.91:8888/download/parser_20190204-c72ca5c0.pkl" -} - - -class ConllCWSReader(object): - """Deprecated. Use ConllLoader for all types of conll-format files.""" - - def __init__(self): - pass - - def load(self, path, cut_long_sent=False): - """ - 返回的DataSet只包含raw_sentence这个field，内容为str。 - 假定了输入为conll的格式，以空行隔开两个句子，每行共7列，即 - :: - - 1 编者按 编者按 NN O 11 nmod:topic - 2 ： ： PU O 11 punct - 3 7月 7月 NT DATE 4 compound:nn - 4 12日 12日 NT DATE 11 nmod:tmod - 5 ， ， PU O 11 punct - - 1 这 这 DT O 3 det - 2 款 款 M O 1 mark:clf - 3 飞行 飞行 NN O 8 nsubj - 4 从 从 P O 5 case - 5 外型 外型 NN O 8 nmod:prep - - """ - datalist = [] - with open(path, 'r', encoding='utf-8') as f: - sample = [] - for line in f: - if line.startswith('\n'): - datalist.append(sample) - sample = [] - elif line.startswith('#'): - continue - else: - sample.append(line.strip().split()) - if len(sample) > 0: - datalist.append(sample) - - ds = DataSet() - for sample in datalist: - # print(sample) - res = self.get_char_lst(sample) - if res is None: - continue - line = ' '.join(res) - if cut_long_sent: - sents = _cut_long_sentence(line) - else: - sents = [line] - for raw_sentence in sents: - ds.append(Instance(raw_sentence=raw_sentence)) - return ds - - def get_char_lst(self, sample): - if len(sample) == 0: - return None - text = [] - for w in sample: - t1, t2, t3, t4 = w[1], w[3], w[6], w[7] - if t3 == '_': - return None - text.append(t1) - return text - - -class ConllxDataLoader(ConllLoader): - """返回“词级别”的标签信息，包括词、词性、（句法）头依赖、（句法）边标签。跟``ZhConllPOSReader``完全不同。 - - Deprecated. Use ConllLoader for all types of conll-format files. - """ - - def __init__(self): - headers = [ - 'words', 'pos_tags', 'heads', 'labels', - ] - indexs = [ - 1, 3, 6, 7, - ] - super(ConllxDataLoader, self).__init__(headers=headers, indexes=indexs) - - -class API: - def __init__(self): - self.pipeline = None - self._dict = None - - def predict(self, *args, **kwargs): - """Do prediction for the given input. - """ - raise NotImplementedError - - def test(self, file_path): - """Test performance over the given data set. - - :param str file_path: - :return: a dictionary of metric values - """ - raise NotImplementedError - - def load(self, path, device): - if os.path.exists(os.path.expanduser(path)): - _dict = torch.load(path, map_location='cpu') - else: - _dict = load_url(path, map_location='cpu') - self._dict = _dict - self.pipeline = _dict['pipeline'] - for processor in self.pipeline.pipeline: - if isinstance(processor, ModelProcessor): - processor.set_model_device(device) - - -class POS(API): - """FastNLP API for Part-Of-Speech tagging. - - :param str model_path: the path to the model. - :param str device: device name such as "cpu" or "cuda:0". Use the same notation as PyTorch. - - """ - - def __init__(self, model_path=None, device='cpu'): - super(POS, self).__init__() - if model_path is None: - model_path = model_urls['pos'] - - self.load(model_path, device) - - def predict(self, content): - """predict函数的介绍， - 函数介绍的第二句，这句话不会换行 - - :param content: list of list of str. Each string is a token(word). - :return answer: list of list of str. Each string is a tag. - """ - if not hasattr(self, "pipeline"): - raise ValueError("You have to load model first.") - - sentence_list = content - # 1. 检查sentence的类型 - for sentence in sentence_list: - if not all((type(obj) == str for obj in sentence)): - raise ValueError("Input must be list of list of string.") - - # 2. 组建dataset - dataset = DataSet() - dataset.add_field("words", sentence_list) - - # 3. 使用pipeline - self.pipeline(dataset) - - def merge_tag(words_list, tags_list): - rtn = [] - for words, tags in zip(words_list, tags_list): - rtn.append([w + "/" + t for w, t in zip(words, tags)]) - return rtn - - output = dataset.field_arrays["tag"].content - if isinstance(content, str): - return output[0] - elif isinstance(content, list): - return merge_tag(content, output) - - def test(self, file_path): - test_data = ConllxDataLoader().load(file_path) - - save_dict = self._dict - tag_vocab = save_dict["tag_vocab"] - pipeline = save_dict["pipeline"] - index_tag = IndexerProcessor(vocab=tag_vocab, field_name="tag", new_added_field_name="truth", is_input=False) - pipeline.pipeline = [index_tag] + pipeline.pipeline - - test_data.rename_field("pos_tags", "tag") - pipeline(test_data) - test_data.set_target("truth") - prediction = test_data.field_arrays["predict"].content - truth = test_data.field_arrays["truth"].content - seq_len = test_data.field_arrays["word_seq_origin_len"].content - - # padding by hand - max_length = max([len(seq) for seq in prediction]) - for idx in range(len(prediction)): - prediction[idx] = list(prediction[idx]) + ([0] * (max_length - len(prediction[idx]))) - truth[idx] = list(truth[idx]) + ([0] * (max_length - len(truth[idx]))) - evaluator = SpanFPreRecMetric(tag_vocab=tag_vocab, pred="predict", target="truth", - seq_len="word_seq_origin_len") - evaluator({"predict": torch.Tensor(prediction), "word_seq_origin_len": torch.Tensor(seq_len)}, - {"truth": torch.Tensor(truth)}) - test_result = evaluator.get_metric() - f1 = round(test_result['f'] * 100, 2) - pre = round(test_result['pre'] * 100, 2) - rec = round(test_result['rec'] * 100, 2) - - return {"F1": f1, "precision": pre, "recall": rec} - - -class CWS(API): - """ - 中文分词高级接口。 - - :param model_path: 当model_path为None，使用默认位置的model。如果默认位置不存在，则自动下载模型 - :param device: str，可以为'cpu', 'cuda'或'cuda:0'等。会将模型load到相应device进行推断。 - """ - - def __init__(self, model_path=None, device='cpu'): - - super(CWS, self).__init__() - if model_path is None: - model_path = model_urls['cws'] - - self.load(model_path, device) - - def predict(self, content): - """ - 分词接口。 - - :param content: str或List[str], 例如: "中文分词很重要！"， 返回的结果是"中文 分词 很 重要 !"。 如果传入的为List[str]，比如 - [ "中文分词很重要！", ...], 返回的结果["中文 分词 很 重要 !", ...]。 - :return: str或List[str], 根据输入的的类型决定。 - """ - if not hasattr(self, 'pipeline'): - raise ValueError("You have to load model first.") - - sentence_list = [] - # 1. 检查sentence的类型 - if isinstance(content, str): - sentence_list.append(content) - elif isinstance(content, list): - sentence_list = content - - # 2. 组建dataset - dataset = DataSet() - dataset.add_field('raw_sentence', sentence_list) - - # 3. 使用pipeline - self.pipeline(dataset) - - output = dataset.get_field('output').content - if isinstance(content, str): - return output[0] - elif isinstance(content, list): - return output - - def test(self, filepath): - """ - 传入一个分词文件路径，返回该数据集上分词f1, precision, recall。 - 分词文件应该为:: - - 1 编者按 编者按 NN O 11 nmod:topic - 2 ： ： PU O 11 punct - 3 7月 7月 NT DATE 4 compound:nn - 4 12日 12日 NT DATE 11 nmod:tmod - 5 ， ， PU O 11 punct - - 1 这 这 DT O 3 det - 2 款 款 M O 1 mark:clf - 3 飞行 飞行 NN O 8 nsubj - 4 从 从 P O 5 case - 5 外型 外型 NN O 8 nmod:prep - - 以空行分割两个句子，有内容的每行有7列。 - - :param filepath: str, 文件路径路径。 - :return: float, float, float. 分别f1, precision, recall. - """ - tag_proc = self._dict['tag_proc'] - cws_model = self.pipeline.pipeline[-2].model - pipeline = self.pipeline.pipeline[:-2] - - pipeline.insert(1, tag_proc) - pp = Pipeline(pipeline) - - reader = ConllCWSReader() - - # te_filename = '/home/hyan/ctb3/test.conllx' - te_dataset = reader.load(filepath) - pp(te_dataset) - - from ..core.tester import Tester - from ..core.metrics import SpanFPreRecMetric - - tester = Tester(data=te_dataset, model=cws_model, metrics=SpanFPreRecMetric(tag_proc.get_vocab()), batch_size=64, - verbose=0) - eval_res = tester.test() - - f1 = eval_res['SpanFPreRecMetric']['f'] - pre = eval_res['SpanFPreRecMetric']['pre'] - rec = eval_res['SpanFPreRecMetric']['rec'] - # print("f1:{:.2f}, pre:{:.2f}, rec:{:.2f}".format(f1, pre, rec)) - - return {"F1": f1, "precision": pre, "recall": rec} - - -class Parser(API): - def __init__(self, model_path=None, device='cpu'): - super(Parser, self).__init__() - if model_path is None: - model_path = model_urls['parser'] - - self.pos_tagger = POS(device=device) - self.load(model_path, device) - - def predict(self, content): - if not hasattr(self, 'pipeline'): - raise ValueError("You have to load model first.") - - # 1. 利用POS得到分词和pos tagging结果 - pos_out = self.pos_tagger.predict(content) - # pos_out = ['这里/NN 是/VB 分词/NN 结果/NN'.split()] - - # 2. 组建dataset - dataset = DataSet() - dataset.add_field('wp', pos_out) - dataset.apply(lambda x: [''] + [w.split('/')[0] for w in x['wp']], new_field_name='words') - dataset.apply(lambda x: [''] + [w.split('/')[1] for w in x['wp']], new_field_name='pos') - dataset.rename_field("words", "raw_words") - - # 3. 使用pipeline - self.pipeline(dataset) - dataset.apply(lambda x: [str(arc) for arc in x['arc_pred']], new_field_name='arc_pred') - dataset.apply(lambda x: [arc + '/' + label for arc, label in - zip(x['arc_pred'], x['label_pred_seq'])][1:], new_field_name='output') - # output like: [['2/top', '0/root', '4/nn', '2/dep']] - return dataset.field_arrays['output'].content - - def load_test_file(self, path): - def get_one(sample): - sample = list(map(list, zip(*sample))) - if len(sample) == 0: - return None - for w in sample[7]: - if w == '_': - print('Error Sample {}'.format(sample)) - return None - # return word_seq, pos_seq, head_seq, head_tag_seq - return sample[1], sample[3], list(map(int, sample[6])), sample[7] - - datalist = [] - with open(path, 'r', encoding='utf-8') as f: - sample = [] - for line in f: - if line.startswith('\n'): - datalist.append(sample) - sample = [] - elif line.startswith('#'): - continue - else: - sample.append(line.split('\t')) - if len(sample) > 0: - datalist.append(sample) - - data = [get_one(sample) for sample in datalist] - data_list = list(filter(lambda x: x is not None, data)) - return data_list - - def test(self, filepath): - data = self.load_test_file(filepath) - - def convert(data): - BOS = '' - dataset = DataSet() - for sample in data: - word_seq = [BOS] + sample[0] - pos_seq = [BOS] + sample[1] - heads = [0] + sample[2] - head_tags = [BOS] + sample[3] - dataset.append(Instance(raw_words=word_seq, - pos=pos_seq, - gold_heads=heads, - arc_true=heads, - tags=head_tags)) - return dataset - - ds = convert(data) - pp = self.pipeline - for p in pp: - if p.field_name == 'word_list': - p.field_name = 'gold_words' - elif p.field_name == 'pos_list': - p.field_name = 'gold_pos' - # ds.rename_field("words", "raw_words") - # ds.rename_field("tag", "pos") - pp(ds) - head_cor, label_cor, total = 0, 0, 0 - for ins in ds: - head_gold = ins['gold_heads'] - head_pred = ins['arc_pred'] - length = len(head_gold) - total += length - for i in range(length): - head_cor += 1 if head_pred[i] == head_gold[i] else 0 - uas = head_cor / total - # print('uas:{:.2f}'.format(uas)) - - for p in pp: - if p.field_name == 'gold_words': - p.field_name = 'word_list' - elif p.field_name == 'gold_pos': - p.field_name = 'pos_list' - - return {"USA": round(uas, 5)} - - -class Analyzer: - def __init__(self, device='cpu'): - - self.cws = CWS(device=device) - self.pos = POS(device=device) - self.parser = Parser(device=device) - - def predict(self, content, seg=False, pos=False, parser=False): - if seg is False and pos is False and parser is False: - seg = True - output_dict = {} - if seg: - seg_output = self.cws.predict(content) - output_dict['seg'] = seg_output - if pos: - pos_output = self.pos.predict(content) - output_dict['pos'] = pos_output - if parser: - parser_output = self.parser.predict(content) - output_dict['parser'] = parser_output - - return output_dict - - def test(self, filepath): - output_dict = {} - if self.cws: - seg_output = self.cws.test(filepath) - output_dict['seg'] = seg_output - if self.pos: - pos_output = self.pos.test(filepath) - output_dict['pos'] = pos_output - if self.parser: - parser_output = self.parser.test(filepath) - output_dict['parser'] = parser_output - - return output_dict diff --git a/legacy/api/converter.py b/legacy/api/converter.py deleted file mode 100644 index 4e03e465..00000000 --- a/legacy/api/converter.py +++ /dev/null @@ -1,181 +0,0 @@ -import re - - -class SpanConverter: - def __init__(self, replace_tag, pattern): - super(SpanConverter, self).__init__() - - self.replace_tag = replace_tag - self.pattern = pattern - - def find_certain_span_and_replace(self, sentence): - replaced_sentence = '' - prev_end = 0 - for match in re.finditer(self.pattern, sentence): - start, end = match.span() - span = sentence[start:end] - replaced_sentence += sentence[prev_end:start] + self.span_to_special_tag(span) - prev_end = end - replaced_sentence += sentence[prev_end:] - - return replaced_sentence - - def span_to_special_tag(self, span): - - return self.replace_tag - - def find_certain_span(self, sentence): - spans = [] - for match in re.finditer(self.pattern, sentence): - spans.append(match.span()) - return spans - - -class AlphaSpanConverter(SpanConverter): - def __init__(self): - replace_tag = '' - # 理想状态下仅处理纯为字母的情况, 但不处理<[a-zA-Z]+>(因为这应该是特殊的tag). - pattern = '[a-zA-Z]+(?=[\u4e00-\u9fff ,%.!<\\-"])' - - super(AlphaSpanConverter, self).__init__(replace_tag, pattern) - - -class DigitSpanConverter(SpanConverter): - def __init__(self): - replace_tag = '' - pattern = '\d[\d\\.]*(?=[\u4e00-\u9fff ,%.!<-])' - - super(DigitSpanConverter, self).__init__(replace_tag, pattern) - - def span_to_special_tag(self, span): - # return self.special_tag - if span[0] == '0' and len(span) > 2: - return '' - decimal_point_count = 0 # one might have more than one decimal pointers - for idx, char in enumerate(span): - if char == '.' or char == '﹒' or char == '·': - decimal_point_count += 1 - if span[-1] == '.' or span[-1] == '﹒' or span[-1] == '·': - # last digit being decimal point means this is not a number - if decimal_point_count == 1: - return span - else: - return '' - if decimal_point_count == 1: - return '' - elif decimal_point_count > 1: - return '' - else: - return '' - - -class TimeConverter(SpanConverter): - def __init__(self): - replace_tag = '' - pattern = '\d+[:：∶][\d:：∶]+(?=[\u4e00-\u9fff ,%.!<-])' - - super().__init__(replace_tag, pattern) - - -class MixNumAlphaConverter(SpanConverter): - def __init__(self): - replace_tag = '' - pattern = None - - super().__init__(replace_tag, pattern) - - def find_certain_span_and_replace(self, sentence): - replaced_sentence = '' - start = 0 - matching_flag = False - number_flag = False - alpha_flag = False - link_flag = False - slash_flag = False - bracket_flag = False - for idx in range(len(sentence)): - if re.match('[0-9a-zA-Z/\\(\\)\'′&\\-]', sentence[idx]): - if not matching_flag: - replaced_sentence += sentence[start:idx] - start = idx - if re.match('[0-9]', sentence[idx]): - number_flag = True - elif re.match('[\'′&\\-]', sentence[idx]): - link_flag = True - elif re.match('/', sentence[idx]): - slash_flag = True - elif re.match('[\\(\\)]', sentence[idx]): - bracket_flag = True - else: - alpha_flag = True - matching_flag = True - elif re.match('[\\.]', sentence[idx]): - pass - else: - if matching_flag: - if (number_flag and alpha_flag) or (link_flag and alpha_flag) \ - or (slash_flag and alpha_flag) or (link_flag and number_flag) \ - or (number_flag and bracket_flag) or (bracket_flag and alpha_flag): - span = sentence[start:idx] - start = idx - replaced_sentence += self.span_to_special_tag(span) - matching_flag = False - number_flag = False - alpha_flag = False - link_flag = False - slash_flag = False - bracket_flag = False - - replaced_sentence += sentence[start:] - return replaced_sentence - - def find_certain_span(self, sentence): - spans = [] - start = 0 - matching_flag = False - number_flag = False - alpha_flag = False - link_flag = False - slash_flag = False - bracket_flag = False - for idx in range(len(sentence)): - if re.match('[0-9a-zA-Z/\\(\\)\'′&\\-]', sentence[idx]): - if not matching_flag: - start = idx - if re.match('[0-9]', sentence[idx]): - number_flag = True - elif re.match('[\'′&\\-]', sentence[idx]): - link_flag = True - elif re.match('/', sentence[idx]): - slash_flag = True - elif re.match('[\\(\\)]', sentence[idx]): - bracket_flag = True - else: - alpha_flag = True - matching_flag = True - elif re.match('[\\.]', sentence[idx]): - pass - else: - if matching_flag: - if (number_flag and alpha_flag) or (link_flag and alpha_flag) \ - or (slash_flag and alpha_flag) or (link_flag and number_flag) \ - or (number_flag and bracket_flag) or (bracket_flag and alpha_flag): - spans.append((start, idx)) - start = idx - - matching_flag = False - number_flag = False - alpha_flag = False - link_flag = False - slash_flag = False - bracket_flag = False - - return spans - - -class EmailConverter(SpanConverter): - def __init__(self): - replaced_tag = "" - pattern = '[0-9a-zA-Z]+[@][.﹒0-9a-zA-Z@]+(?=[\u4e00-\u9fff ,%.!<\\-"$])' - - super(EmailConverter, self).__init__(replaced_tag, pattern) diff --git a/legacy/api/examples.py b/legacy/api/examples.py deleted file mode 100644 index c1b2e155..00000000 --- a/legacy/api/examples.py +++ /dev/null @@ -1,56 +0,0 @@ -""" -api/example.py contains all API examples provided by fastNLP. -It is used as a tutorial for API or a test script since it is difficult to test APIs in travis. - -""" -from . import CWS, POS, Parser - -text = ['编者按：7月12日，英国航空航天系统公司公布了该公司研制的第一款高科技隐形无人机雷电之神。', - '这款飞行从外型上来看酷似电影中的太空飞行器，据英国方面介绍，可以实现洲际远程打击。', - '那么这款无人机到底有多厉害？'] - - -def chinese_word_segmentation(): - cws = CWS(device='cpu') - print(cws.predict(text)) - - -def chinese_word_segmentation_test(): - cws = CWS(device='cpu') - print(cws.test("../../test/data_for_tests/zh_sample.conllx")) - - -def pos_tagging(): - # 输入已分词序列 - text = [['编者', '按：', '7月', '12日', '，', '英国', '航空', '航天', '系统', '公司', '公布', '了', '该', '公司', - '研制', '的', '第一款', '高科技', '隐形', '无人机', '雷电之神', '。'], - ['那么', '这', '款', '无人机', '到底', '有', '多', '厉害', '？']] - pos = POS(device='cpu') - print(pos.predict(text)) - - -def pos_tagging_test(): - pos = POS(device='cpu') - print(pos.test("../../test/data_for_tests/zh_sample.conllx")) - - -def syntactic_parsing(): - text = [['编者', '按：', '7月', '12日', '，', '英国', '航空', '航天', '系统', '公司', '公布', '了', '该', '公司', - '研制', '的', '第一款', '高科技', '隐形', '无人机', '雷电之神', '。'], - ['那么', '这', '款', '无人机', '到底', '有', '多', '厉害', '？']] - parser = Parser(device='cpu') - print(parser.predict(text)) - - -def syntactic_parsing_test(): - parser = Parser(device='cpu') - print(parser.test("../../test/data_for_tests/zh_sample.conllx")) - - -if __name__ == "__main__": - # chinese_word_segmentation() - # chinese_word_segmentation_test() - # pos_tagging() - # pos_tagging_test() - syntactic_parsing() - # syntactic_parsing_test() diff --git a/legacy/api/pipeline.py b/legacy/api/pipeline.py deleted file mode 100644 index 2cec16b3..00000000 --- a/legacy/api/pipeline.py +++ /dev/null @@ -1,33 +0,0 @@ -from ..api.processor import Processor - - -class Pipeline: - """ - Pipeline takes a DataSet object as input, runs multiple processors sequentially, and - outputs a DataSet object. - """ - - def __init__(self, processors=None): - self.pipeline = [] - if isinstance(processors, list): - for proc in processors: - assert isinstance(proc, Processor), "Must be a Processor, not {}.".format(type(proc)) - self.pipeline = processors - - def add_processor(self, processor): - assert isinstance(processor, Processor), "Must be a Processor, not {}.".format(type(processor)) - self.pipeline.append(processor) - - def process(self, dataset): - assert len(self.pipeline) != 0, "You need to add some processor first." - - for proc in self.pipeline: - dataset = proc(dataset) - - return dataset - - def __call__(self, *args, **kwargs): - return self.process(*args, **kwargs) - - def __getitem__(self, item): - return self.pipeline[item] diff --git a/legacy/api/processor.py b/legacy/api/processor.py deleted file mode 100644 index 4c442ed2..00000000 --- a/legacy/api/processor.py +++ /dev/null @@ -1,428 +0,0 @@ -import re -from collections import defaultdict - -import torch - -from fastNLP.core.batch import Batch -from fastNLP.core.dataset import DataSet -from fastNLP.core.sampler import SequentialSampler -from fastNLP.core.vocabulary import Vocabulary - - -class Processor(object): - def __init__(self, field_name, new_added_field_name): - """ - - :param field_name: 处理哪个field - :param new_added_field_name: 如果为None，则认为是field_name，即覆盖原有的field - """ - self.field_name = field_name - if new_added_field_name is None: - self.new_added_field_name = field_name - else: - self.new_added_field_name = new_added_field_name - - def process(self, *args, **kwargs): - raise NotImplementedError - - def __call__(self, *args, **kwargs): - return self.process(*args, **kwargs) - - -class FullSpaceToHalfSpaceProcessor(Processor): - """全角转半角，以字符为处理单元 - - """ - - def __init__(self, field_name, change_alpha=True, change_digit=True, change_punctuation=True, - change_space=True): - super(FullSpaceToHalfSpaceProcessor, self).__init__(field_name, None) - - self.change_alpha = change_alpha - self.change_digit = change_digit - self.change_punctuation = change_punctuation - self.change_space = change_space - - FH_SPACE = [(u"　", u" ")] - FH_NUM = [ - (u"０", u"0"), (u"１", u"1"), (u"２", u"2"), (u"３", u"3"), (u"４", u"4"), - (u"５", u"5"), (u"６", u"6"), (u"７", u"7"), (u"８", u"8"), (u"９", u"9")] - FH_ALPHA = [ - (u"ａ", u"a"), (u"ｂ", u"b"), (u"ｃ", u"c"), (u"ｄ", u"d"), (u"ｅ", u"e"), - (u"ｆ", u"f"), (u"ｇ", u"g"), (u"ｈ", u"h"), (u"ｉ", u"i"), (u"ｊ", u"j"), - (u"ｋ", u"k"), (u"ｌ", u"l"), (u"ｍ", u"m"), (u"ｎ", u"n"), (u"ｏ", u"o"), - (u"ｐ", u"p"), (u"ｑ", u"q"), (u"ｒ", u"r"), (u"ｓ", u"s"), (u"ｔ", u"t"), - (u"ｕ", u"u"), (u"ｖ", u"v"), (u"ｗ", u"w"), (u"ｘ", u"x"), (u"ｙ", u"y"), - (u"ｚ", u"z"), - (u"Ａ", u"A"), (u"Ｂ", u"B"), (u"Ｃ", u"C"), (u"Ｄ", u"D"), (u"Ｅ", u"E"), - (u"Ｆ", u"F"), (u"Ｇ", u"G"), (u"Ｈ", u"H"), (u"Ｉ", u"I"), (u"Ｊ", u"J"), - (u"Ｋ", u"K"), (u"Ｌ", u"L"), (u"Ｍ", u"M"), (u"Ｎ", u"N"), (u"Ｏ", u"O"), - (u"Ｐ", u"P"), (u"Ｑ", u"Q"), (u"Ｒ", u"R"), (u"Ｓ", u"S"), (u"Ｔ", u"T"), - (u"Ｕ", u"U"), (u"Ｖ", u"V"), (u"Ｗ", u"W"), (u"Ｘ", u"X"), (u"Ｙ", u"Y"), - (u"Ｚ", u"Z")] - # 谨慎使用标点符号转换, 因为"5．12特大地震"转换后可能就成了"5.12特大地震" - FH_PUNCTUATION = [ - (u'％', u'%'), (u'！', u'!'), (u'＂', u'\"'), (u'＇', u'\''), (u'＃', u'#'), - (u'￥', u'$'), (u'＆', u'&'), (u'（', u'('), (u'）', u')'), (u'＊', u'*'), - (u'＋', u'+'), (u'，', u','), (u'－', u'-'), (u'．', u'.'), (u'／', u'/'), - (u'：', u':'), (u'；', u';'), (u'＜', u'<'), (u'＝', u'='), (u'＞', u'>'), - (u'？', u'?'), (u'＠', u'@'), (u'［', u'['), (u'］', u']'), (u'＼', u'\\'), - (u'＾', u'^'), (u'＿', u'_'), (u'｀', u'`'), (u'～', u'~'), (u'｛', u'{'), - (u'｝', u'}'), (u'｜', u'|')] - FHs = [] - if self.change_alpha: - FHs = FH_ALPHA - if self.change_digit: - FHs += FH_NUM - if self.change_punctuation: - FHs += FH_PUNCTUATION - if self.change_space: - FHs += FH_SPACE - self.convert_map = {k: v for k, v in FHs} - - def process(self, dataset): - assert isinstance(dataset, DataSet), "Only Dataset class is allowed, not {}.".format(type(dataset)) - - def inner_proc(ins): - sentence = ins[self.field_name] - new_sentence = [""] * len(sentence) - for idx, char in enumerate(sentence): - if char in self.convert_map: - char = self.convert_map[char] - new_sentence[idx] = char - return "".join(new_sentence) - - dataset.apply(inner_proc, new_field_name=self.field_name) - return dataset - - -class PreAppendProcessor(Processor): - """ - 向某个field的起始增加data(应该为str类型)。该field需要为list类型。即新增的field为 - [data] + instance[field_name] - - """ - - def __init__(self, data, field_name, new_added_field_name=None): - super(PreAppendProcessor, self).__init__(field_name, new_added_field_name) - self.data = data - - def process(self, dataset): - dataset.apply(lambda ins: [self.data] + ins[self.field_name], new_field_name=self.new_added_field_name) - return dataset - - -class SliceProcessor(Processor): - """ - 从某个field中只取部分内容。等价于instance[field_name][start:end:step] - - """ - - def __init__(self, start, end, step, field_name, new_added_field_name=None): - super(SliceProcessor, self).__init__(field_name, new_added_field_name) - for o in (start, end, step): - assert isinstance(o, int) or o is None - self.slice = slice(start, end, step) - - def process(self, dataset): - dataset.apply(lambda ins: ins[self.field_name][self.slice], new_field_name=self.new_added_field_name) - return dataset - - -class Num2TagProcessor(Processor): - """ - 将一句话中的数字转换为某个tag。 - - """ - - def __init__(self, tag, field_name, new_added_field_name=None): - """ - - :param tag: str, 将数字转换为该tag - :param field_name: - :param new_added_field_name: - """ - super(Num2TagProcessor, self).__init__(field_name, new_added_field_name) - self.tag = tag - self.pattern = r'[-+]?([0-9]+[.]?[0-9]*)+[/eE]?[-+]?([0-9]+[.]?[0-9]*)' - - def process(self, dataset): - - def inner_proc(ins): - s = ins[self.field_name] - new_s = [None] * len(s) - for i, w in enumerate(s): - if re.search(self.pattern, w) is not None: - w = self.tag - new_s[i] = w - return new_s - - dataset.apply(inner_proc, new_field_name=self.new_added_field_name) - return dataset - - -class IndexerProcessor(Processor): - """ - 给定一个vocabulary , 将指定field转换为index形式。指定field应该是一维的list，比如 - ['我', '是', xxx] - """ - - def __init__(self, vocab, field_name, new_added_field_name, delete_old_field=False, is_input=True): - - assert isinstance(vocab, Vocabulary), "Only Vocabulary class is allowed, not {}.".format(type(vocab)) - - super(IndexerProcessor, self).__init__(field_name, new_added_field_name) - self.vocab = vocab - self.delete_old_field = delete_old_field - self.is_input = is_input - - def set_vocab(self, vocab): - assert isinstance(vocab, Vocabulary), "Only Vocabulary class is allowed, not {}.".format(type(vocab)) - - self.vocab = vocab - - def process(self, dataset): - assert isinstance(dataset, DataSet), "Only DataSet class is allowed, not {}.".format(type(dataset)) - dataset.apply(lambda ins: [self.vocab.to_index(token) for token in ins[self.field_name]], - new_field_name=self.new_added_field_name) - if self.is_input: - dataset.set_input(self.new_added_field_name) - - if self.delete_old_field: - dataset.delete_field(self.field_name) - - return dataset - - -class VocabProcessor(Processor): - """ - 传入若干个DataSet以建立vocabulary。 - - """ - - def __init__(self, field_name, min_freq=1, max_size=None): - super(VocabProcessor, self).__init__(field_name, None) - self.vocab = Vocabulary(min_freq=min_freq, max_size=max_size) - - def process(self, *datasets): - for dataset in datasets: - assert isinstance(dataset, DataSet), "Only Dataset class is allowed, not {}.".format(type(dataset)) - dataset.apply(lambda ins: self.vocab.update(ins[self.field_name])) - - def get_vocab(self): - self.vocab.build_vocab() - return self.vocab - - -class SeqLenProcessor(Processor): - """ - 根据某个field新增一个sequence length的field。取该field的第一维 - - """ - - def __init__(self, field_name, new_added_field_name='seq_lens', is_input=True): - super(SeqLenProcessor, self).__init__(field_name, new_added_field_name) - self.is_input = is_input - - def process(self, dataset): - assert isinstance(dataset, DataSet), "Only Dataset class is allowed, not {}.".format(type(dataset)) - dataset.apply(lambda ins: len(ins[self.field_name]), new_field_name=self.new_added_field_name) - if self.is_input: - dataset.set_input(self.new_added_field_name) - return dataset - - -from fastNLP.core.utils import _build_args - - -class ModelProcessor(Processor): - def __init__(self, model, seq_len_field_name='seq_lens', batch_size=32): - """ - 传入一个model，在process()时传入一个dataset，该processor会通过Batch将DataSet的内容输出给model.predict或者model.forward. - model输出的内容会被增加到dataset中，field_name由model输出决定。如果生成的内容维度不是(Batch_size, )与 - (Batch_size, 1)，则使用seqence length这个field进行unpad - TODO 这个类需要删除对seq_lens的依赖。 - - :param seq_len_field_name: - :param batch_size: - """ - super(ModelProcessor, self).__init__(None, None) - self.batch_size = batch_size - self.seq_len_field_name = seq_len_field_name - self.model = model - - def process(self, dataset): - self.model.eval() - assert isinstance(dataset, DataSet), "Only Dataset class is allowed, not {}.".format(type(dataset)) - data_iterator = Batch(dataset, batch_size=self.batch_size, sampler=SequentialSampler()) - - batch_output = defaultdict(list) - predict_func = self.model.forward - with torch.no_grad(): - for batch_x, _ in data_iterator: - refined_batch_x = _build_args(predict_func, **batch_x) - prediction = predict_func(**refined_batch_x) - seq_lens = batch_x[self.seq_len_field_name].tolist() - - for key, value in prediction.items(): - tmp_batch = [] - value = value.cpu().numpy() - if len(value.shape) == 1 or (len(value.shape) == 2 and value.shape[1] == 1): - batch_output[key].extend(value.tolist()) - else: - for idx, seq_len in enumerate(seq_lens): - tmp_batch.append(value[idx, :seq_len]) - batch_output[key].extend(tmp_batch) - if not self.seq_len_field_name in prediction: - batch_output[self.seq_len_field_name].extend(seq_lens) - - # TODO 当前的实现会导致之后的processor需要知道model输出的output的key是什么 - for field_name, fields in batch_output.items(): - dataset.add_field(field_name, fields, is_input=True, is_target=False) - - return dataset - - def set_model(self, model): - self.model = model - - def set_model_device(self, device): - device = torch.device(device) - self.model.to(device) - - -class Index2WordProcessor(Processor): - """ - 将DataSet中某个为index的field根据vocab转换为str - - """ - - def __init__(self, vocab, field_name, new_added_field_name): - super(Index2WordProcessor, self).__init__(field_name, new_added_field_name) - self.vocab = vocab - - def process(self, dataset): - dataset.apply(lambda ins: [self.vocab.to_word(w) for w in ins[self.field_name]], - new_field_name=self.new_added_field_name) - return dataset - - -class SetTargetProcessor(Processor): - def __init__(self, *fields, flag=True): - super(SetTargetProcessor, self).__init__(None, None) - self.fields = fields - self.flag = flag - - def process(self, dataset): - dataset.set_target(*self.fields, flag=self.flag) - return dataset - - -class SetInputProcessor(Processor): - def __init__(self, *fields, flag=True): - super(SetInputProcessor, self).__init__(None, None) - self.fields = fields - self.flag = flag - - def process(self, dataset): - dataset.set_input(*self.fields, flag=self.flag) - return dataset - - -class VocabIndexerProcessor(Processor): - """ - 根据DataSet创建Vocabulary，并将其用数字index。新生成的index的field会被放在new_added_filed_name, 如果没有提供 - new_added_field_name, 则覆盖原有的field_name. - - """ - - def __init__(self, field_name, new_added_filed_name=None, min_freq=1, max_size=None, - verbose=0, is_input=True): - """ - - :param field_name: 从哪个field_name创建词表，以及对哪个field_name进行index操作 - :param new_added_filed_name: index时，生成的index field的名称，如果不传入，则覆盖field_name. - :param min_freq: 创建的Vocabulary允许的单词最少出现次数. - :param max_size: 创建的Vocabulary允许的最大的单词数量 - :param verbose: 0, 不输出任何信息；1，输出信息 - :param bool is_input: - """ - super(VocabIndexerProcessor, self).__init__(field_name, new_added_filed_name) - self.min_freq = min_freq - self.max_size = max_size - - self.verbose = verbose - self.is_input = is_input - - def construct_vocab(self, *datasets): - """ - 使用传入的DataSet创建vocabulary - - :param datasets: DataSet类型的数据，用于构建vocabulary - :return: - """ - self.vocab = Vocabulary(min_freq=self.min_freq, max_size=self.max_size) - for dataset in datasets: - assert isinstance(dataset, DataSet), "Only Dataset class is allowed, not {}.".format(type(dataset)) - dataset.apply(lambda ins: self.vocab.update(ins[self.field_name])) - self.vocab.build_vocab() - if self.verbose: - print("Vocabulary Constructed, has {} items.".format(len(self.vocab))) - - def process(self, *datasets, only_index_dataset=None): - """ - 若还未建立Vocabulary，则使用dataset中的DataSet建立vocabulary；若已经有了vocabulary则使用已有的vocabulary。得到vocabulary - 后，则会index datasets与only_index_dataset。 - - :param datasets: DataSet类型的数据 - :param only_index_dataset: DataSet, or list of DataSet. 该参数中的内容只会被用于index，不会被用于生成vocabulary。 - :return: - """ - if len(datasets) == 0 and not hasattr(self, 'vocab'): - raise RuntimeError("You have to construct vocabulary first. Or you have to pass datasets to construct it.") - if not hasattr(self, 'vocab'): - self.construct_vocab(*datasets) - else: - if self.verbose: - print("Using constructed vocabulary with {} items.".format(len(self.vocab))) - to_index_datasets = [] - if len(datasets) != 0: - for dataset in datasets: - assert isinstance(dataset, DataSet), "Only DataSet class is allowed, not {}.".format(type(dataset)) - to_index_datasets.append(dataset) - - if not (only_index_dataset is None): - if isinstance(only_index_dataset, list): - for dataset in only_index_dataset: - assert isinstance(dataset, DataSet), "Only DataSet class is allowed, not {}.".format(type(dataset)) - to_index_datasets.append(dataset) - elif isinstance(only_index_dataset, DataSet): - to_index_datasets.append(only_index_dataset) - else: - raise TypeError('Only DataSet or list of DataSet is allowed, not {}.'.format(type(only_index_dataset))) - - for dataset in to_index_datasets: - assert isinstance(dataset, DataSet), "Only DataSet class is allowed, not {}.".format(type(dataset)) - dataset.apply(lambda ins: [self.vocab.to_index(token) for token in ins[self.field_name]], - new_field_name=self.new_added_field_name, is_input=self.is_input) - # 只返回一个，infer时为了跟其他processor保持一致 - if len(to_index_datasets) == 1: - return to_index_datasets[0] - - def set_vocab(self, vocab): - assert isinstance(vocab, Vocabulary), "Only fastNLP.core.Vocabulary is allowed, not {}.".format(type(vocab)) - self.vocab = vocab - - def delete_vocab(self): - del self.vocab - - def get_vocab_size(self): - return len(self.vocab) - - def set_verbose(self, verbose): - """ - 设置processor verbose状态。 - - :param verbose: int, 0，不输出任何信息；1，输出vocab 信息。 - :return: - """ - self.verbose = verbose diff --git a/legacy/api/utils.py b/legacy/api/utils.py deleted file mode 100644 index 184e5fe6..00000000 --- a/legacy/api/utils.py +++ /dev/null @@ -1,134 +0,0 @@ -import hashlib -import os -import re -import shutil -import sys -import tempfile - -import torch - -try: - from requests.utils import urlparse - from requests import get as urlopen - requests_available = True -except ImportError: - requests_available = False - if sys.version_info[0] == 2: - from urlparse import urlparse # noqa f811 - from urllib2 import urlopen # noqa f811 - else: - from urllib.request import urlopen - from urllib.parse import urlparse -try: - from tqdm.auto import tqdm -except: - from fastNLP.core.utils import _pseudo_tqdm as tqdm - -# matches bfd8deac from resnet18-bfd8deac.pth -HASH_REGEX = re.compile(r'-([a-f0-9]*)\.') - - -def load_url(url, model_dir=None, map_location=None, progress=True): - r"""Loads the Torch serialized object at the given URL. - - If the object is already present in `model_dir`, it's deserialized and - returned. The filename part of the URL should follow the naming convention - ``filename-.ext`` where ```` is the first eight or more - digits of the SHA256 hash of the contents of the file. The hash is used to - ensure unique names and to verify the contents of the file. - - The default value of `model_dir` is ``$TORCH_HOME/models`` where - ``$TORCH_HOME`` defaults to ``~/.torch``. The default directory can be - overridden with the ``$TORCH_MODEL_ZOO`` environment variable. - - Args: - url (string): URL of the object to download - model_dir (string, optional): directory in which to save the object - map_location (optional): a function or a dict specifying how to remap storage locations (see torch.load) - progress (bool, optional): whether or not to display a progress bar to stderr - - Example: - # >>> state_dict = model_zoo.load_url('https://s3.amazonaws.com/pytorch/models/resnet18-5c106cde.pth') - - """ - if model_dir is None: - torch_home = os.path.expanduser(os.getenv('fastNLP_HOME', '~/.fastNLP')) - model_dir = os.getenv('fastNLP_MODEL_ZOO', os.path.join(torch_home, 'models')) - if not os.path.exists(model_dir): - os.makedirs(model_dir) - parts = urlparse(url) - filename = os.path.basename(parts.path) - cached_file = os.path.join(model_dir, filename) - if not os.path.exists(cached_file): - sys.stderr.write('Downloading: "{}" to {}\n'.format(url, cached_file)) - # hash_prefix = HASH_REGEX.search(filename).group(1) - _download_url_to_file(url, cached_file, hash_prefix=None, progress=progress) - return torch.load(cached_file, map_location=map_location) - - -def _download_url_to_file(url, dst, hash_prefix, progress): - if requests_available: - u = urlopen(url, stream=True) - file_size = int(u.headers["Content-Length"]) - u = u.raw - else: - u = urlopen(url) - meta = u.info() - if hasattr(meta, 'getheaders'): - file_size = int(meta.getheaders("Content-Length")[0]) - else: - file_size = int(meta.get_all("Content-Length")[0]) - - f = tempfile.NamedTemporaryFile(delete=False) - try: - if hash_prefix is not None: - sha256 = hashlib.sha256() - with tqdm(total=file_size, disable=not progress) as pbar: - while True: - buffer = u.read(8192) - if len(buffer) == 0: - break - f.write(buffer) - if hash_prefix is not None: - sha256.update(buffer) - pbar.update(len(buffer)) - - f.close() - if hash_prefix is not None: - digest = sha256.hexdigest() - if digest[:len(hash_prefix)] != hash_prefix: - raise RuntimeError('invalid hash value (expected "{}", got "{}")' - .format(hash_prefix, digest)) - shutil.move(f.name, dst) - finally: - f.close() - if os.path.exists(f.name): - os.remove(f.name) - - -if tqdm is None: - # fake tqdm if it's not installed - class tqdm(object): - - def __init__(self, total, disable=False): - self.total = total - self.disable = disable - self.n = 0 - - def update(self, n): - if self.disable: - return - - self.n += n - sys.stderr.write("\r{0:.1f}%".format(100 * self.n / float(self.total))) - sys.stderr.flush() - - def __enter__(self): - return self - - def __exit__(self, exc_type, exc_val, exc_tb): - if self.disable: - return - - sys.stderr.write('\n') - diff --git a/legacy/automl/__init__.py b/legacy/automl/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/legacy/automl/enas_controller.py b/legacy/automl/enas_controller.py deleted file mode 100644 index 6ddbb211..00000000 --- a/legacy/automl/enas_controller.py +++ /dev/null @@ -1,223 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch -"""A module with NAS controller-related code.""" -import collections -import os - -import torch -import torch.nn.functional as F - -import fastNLP.automl.enas_utils as utils -from fastNLP.automl.enas_utils import Node - - -def _construct_dags(prev_nodes, activations, func_names, num_blocks): - """Constructs a set of DAGs based on the actions, i.e., previous nodes and - activation functions, sampled from the controller/policy pi. - - Args: - prev_nodes: Previous node actions from the policy. - activations: Activations sampled from the policy. - func_names: Mapping from activation function names to functions. - num_blocks: Number of blocks in the target RNN cell. - - Returns: - A list of DAGs defined by the inputs. - - RNN cell DAGs are represented in the following way: - - 1. Each element (node) in a DAG is a list of `Node`s. - - 2. The `Node`s in the list dag[i] correspond to the subsequent nodes - that take the output from node i as their own input. - - 3. dag[-1] is the node that takes input from x^{(t)} and h^{(t - 1)}. - dag[-1] always feeds dag[0]. - dag[-1] acts as if `w_xc`, `w_hc`, `w_xh` and `w_hh` are its - weights. - - 4. dag[N - 1] is the node that produces the hidden state passed to - the next timestep. dag[N - 1] is also always a leaf node, and therefore - is always averaged with the other leaf nodes and fed to the output - decoder. - """ - dags = [] - for nodes, func_ids in zip(prev_nodes, activations): - dag = collections.defaultdict(list) - - # add first node - dag[-1] = [Node(0, func_names[func_ids[0]])] - dag[-2] = [Node(0, func_names[func_ids[0]])] - - # add following nodes - for jdx, (idx, func_id) in enumerate(zip(nodes, func_ids[1:])): - dag[utils.to_item(idx)].append(Node(jdx + 1, func_names[func_id])) - - leaf_nodes = set(range(num_blocks)) - dag.keys() - - # merge with avg - for idx in leaf_nodes: - dag[idx] = [Node(num_blocks, 'avg')] - - # This is actually y^{(t)}. h^{(t)} is node N - 1 in - # the graph, where N Is the number of nodes. I.e., h^{(t)} takes - # only one other node as its input. - # last h[t] node - last_node = Node(num_blocks + 1, 'h[t]') - dag[num_blocks] = [last_node] - dags.append(dag) - - return dags - - -class Controller(torch.nn.Module): - """Based on - https://github.com/pytorch/examples/blob/master/word_language_model/model.py - - RL controllers do not necessarily have much to do with - language models. - - Base the controller RNN on the GRU from: - https://github.com/ikostrikov/pytorch-a2c-ppo-acktr/blob/master/model.py - """ - def __init__(self, num_blocks=4, controller_hid=100, cuda=False): - torch.nn.Module.__init__(self) - - # `num_tokens` here is just the activation function - # for every even step, - self.shared_rnn_activations = ['tanh', 'ReLU', 'identity', 'sigmoid'] - self.num_tokens = [len(self.shared_rnn_activations)] - self.controller_hid = controller_hid - self.use_cuda = cuda - self.num_blocks = num_blocks - for idx in range(num_blocks): - self.num_tokens += [idx + 1, len(self.shared_rnn_activations)] - self.func_names = self.shared_rnn_activations - - num_total_tokens = sum(self.num_tokens) - - self.encoder = torch.nn.Embedding(num_total_tokens, - controller_hid) - self.lstm = torch.nn.LSTMCell(controller_hid, controller_hid) - - # Perhaps these weights in the decoder should be - # shared? At least for the activation functions, which all have the - # same size. - self.decoders = [] - for idx, size in enumerate(self.num_tokens): - decoder = torch.nn.Linear(controller_hid, size) - self.decoders.append(decoder) - - self._decoders = torch.nn.ModuleList(self.decoders) - - self.reset_parameters() - self.static_init_hidden = utils.keydefaultdict(self.init_hidden) - - def _get_default_hidden(key): - return utils.get_variable( - torch.zeros(key, self.controller_hid), - self.use_cuda, - requires_grad=False) - - self.static_inputs = utils.keydefaultdict(_get_default_hidden) - - def reset_parameters(self): - init_range = 0.1 - for param in self.parameters(): - param.data.uniform_(-init_range, init_range) - for decoder in self.decoders: - decoder.bias.data.fill_(0) - - def forward(self, # pylint:disable=arguments-differ - inputs, - hidden, - block_idx, - is_embed): - if not is_embed: - embed = self.encoder(inputs) - else: - embed = inputs - - hx, cx = self.lstm(embed, hidden) - logits = self.decoders[block_idx](hx) - - logits /= 5.0 - - # # exploration - # if self.args.mode == 'train': - # logits = (2.5 * F.tanh(logits)) - - return logits, (hx, cx) - - def sample(self, batch_size=1, with_details=False, save_dir=None): - """Samples a set of `args.num_blocks` many computational nodes from the - controller, where each node is made up of an activation function, and - each node except the last also includes a previous node. - """ - if batch_size < 1: - raise Exception(f'Wrong batch_size: {batch_size} < 1') - - # [B, L, H] - inputs = self.static_inputs[batch_size] - hidden = self.static_init_hidden[batch_size] - - activations = [] - entropies = [] - log_probs = [] - prev_nodes = [] - # The RNN controller alternately outputs an activation, - # followed by a previous node, for each block except the last one, - # which only gets an activation function. The last node is the output - # node, and its previous node is the average of all leaf nodes. - for block_idx in range(2*(self.num_blocks - 1) + 1): - logits, hidden = self.forward(inputs, - hidden, - block_idx, - is_embed=(block_idx == 0)) - - probs = F.softmax(logits, dim=-1) - log_prob = F.log_softmax(logits, dim=-1) - # .mean() for entropy? - entropy = -(log_prob * probs).sum(1, keepdim=False) - - action = probs.multinomial(num_samples=1).data - selected_log_prob = log_prob.gather( - 1, utils.get_variable(action, requires_grad=False)) - - # why the [:, 0] here? Should it be .squeeze(), or - # .view()? Same below with `action`. - entropies.append(entropy) - log_probs.append(selected_log_prob[:, 0]) - - # 0: function, 1: previous node - mode = block_idx % 2 - inputs = utils.get_variable( - action[:, 0] + sum(self.num_tokens[:mode]), - requires_grad=False) - - if mode == 0: - activations.append(action[:, 0]) - elif mode == 1: - prev_nodes.append(action[:, 0]) - - prev_nodes = torch.stack(prev_nodes).transpose(0, 1) - activations = torch.stack(activations).transpose(0, 1) - - dags = _construct_dags(prev_nodes, - activations, - self.func_names, - self.num_blocks) - - if save_dir is not None: - for idx, dag in enumerate(dags): - utils.draw_network(dag, - os.path.join(save_dir, f'graph{idx}.png')) - - if with_details: - return dags, torch.cat(log_probs), torch.cat(entropies) - - return dags - - def init_hidden(self, batch_size): - zeros = torch.zeros(batch_size, self.controller_hid) - return (utils.get_variable(zeros, self.use_cuda, requires_grad=False), - utils.get_variable(zeros.clone(), self.use_cuda, requires_grad=False)) diff --git a/legacy/automl/enas_model.py b/legacy/automl/enas_model.py deleted file mode 100644 index 4f9fb449..00000000 --- a/legacy/automl/enas_model.py +++ /dev/null @@ -1,388 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch - -"""Module containing the shared RNN model.""" -import collections - -import numpy as np -import torch -import torch.nn.functional as F -from torch import nn -from torch.autograd import Variable - -import fastNLP.automl.enas_utils as utils -from fastNLP.models.base_model import BaseModel - - -def _get_dropped_weights(w_raw, dropout_p, is_training): - """Drops out weights to implement DropConnect. - - Args: - w_raw: Full, pre-dropout, weights to be dropped out. - dropout_p: Proportion of weights to drop out. - is_training: True iff _shared_ model is training. - - Returns: - The dropped weights. - - Why does torch.nn.functional.dropout() return: - 1. `torch.autograd.Variable()` on the training loop - 2. `torch.nn.Parameter()` on the controller or eval loop, when - training = False... - - Even though the call to `_setweights` in the Smerity repo's - `weight_drop.py` does not have this behaviour, and `F.dropout` always - returns `torch.autograd.Variable` there, even when `training=False`? - - The above TODO is the reason for the hacky check for `torch.nn.Parameter`. - """ - dropped_w = F.dropout(w_raw, p=dropout_p, training=is_training) - - if isinstance(dropped_w, torch.nn.Parameter): - dropped_w = dropped_w.clone() - - return dropped_w - -class EmbeddingDropout(torch.nn.Embedding): - """Class for dropping out embeddings by zero'ing out parameters in the - embedding matrix. - - This is equivalent to dropping out particular words, e.g., in the sentence - 'the quick brown fox jumps over the lazy dog', dropping out 'the' would - lead to the sentence '### quick brown fox jumps over ### lazy dog' (in the - embedding vector space). - - See 'A Theoretically Grounded Application of Dropout in Recurrent Neural - Networks', (Gal and Ghahramani, 2016). - """ - def __init__(self, - num_embeddings, - embedding_dim, - max_norm=None, - norm_type=2, - scale_grad_by_freq=False, - sparse=False, - dropout=0.1, - scale=None): - """Embedding constructor. - - Args: - dropout: Dropout probability. - scale: Used to scale parameters of embedding weight matrix that are - not dropped out. Note that this is _in addition_ to the - `1/(1 - dropout)` scaling. - - See `torch.nn.Embedding` for remaining arguments. - """ - torch.nn.Embedding.__init__(self, - num_embeddings=num_embeddings, - embedding_dim=embedding_dim, - max_norm=max_norm, - norm_type=norm_type, - scale_grad_by_freq=scale_grad_by_freq, - sparse=sparse) - self.dropout = dropout - assert (dropout >= 0.0) and (dropout < 1.0), ('Dropout must be >= 0.0 ' - 'and < 1.0') - self.scale = scale - - def forward(self, inputs): # pylint:disable=arguments-differ - """Embeds `inputs` with the dropped out embedding weight matrix.""" - if self.training: - dropout = self.dropout - else: - dropout = 0 - - if dropout: - mask = self.weight.data.new(self.weight.size(0), 1) - mask.bernoulli_(1 - dropout) - mask = mask.expand_as(self.weight) - mask = mask / (1 - dropout) - masked_weight = self.weight * Variable(mask) - else: - masked_weight = self.weight - if self.scale and self.scale != 1: - masked_weight = masked_weight * self.scale - - return F.embedding(inputs, - masked_weight, - max_norm=self.max_norm, - norm_type=self.norm_type, - scale_grad_by_freq=self.scale_grad_by_freq, - sparse=self.sparse) - - -class LockedDropout(nn.Module): - # code from https://github.com/salesforce/awd-lstm-lm/blob/master/locked_dropout.py - def __init__(self): - super().__init__() - - def forward(self, x, dropout=0.5): - if not self.training or not dropout: - return x - m = x.data.new(1, x.size(1), x.size(2)).bernoulli_(1 - dropout) - mask = Variable(m, requires_grad=False) / (1 - dropout) - mask = mask.expand_as(x) - return mask * x - - -class ENASModel(BaseModel): - """Shared RNN model.""" - def __init__(self, embed_num, num_classes, num_blocks=4, cuda=False, shared_hid=1000, shared_embed=1000): - super(ENASModel, self).__init__() - - self.use_cuda = cuda - - self.shared_hid = shared_hid - self.num_blocks = num_blocks - self.decoder = nn.Linear(self.shared_hid, num_classes) - self.encoder = EmbeddingDropout(embed_num, - shared_embed, - dropout=0.1) - self.lockdrop = LockedDropout() - self.dag = None - - # Tie weights - # self.decoder.weight = self.encoder.weight - - # Since W^{x, c} and W^{h, c} are always summed, there - # is no point duplicating their bias offset parameter. Likewise for - # W^{x, h} and W^{h, h}. - self.w_xc = nn.Linear(shared_embed, self.shared_hid) - self.w_xh = nn.Linear(shared_embed, self.shared_hid) - - # The raw weights are stored here because the hidden-to-hidden weights - # are weight dropped on the forward pass. - self.w_hc_raw = torch.nn.Parameter( - torch.Tensor(self.shared_hid, self.shared_hid)) - self.w_hh_raw = torch.nn.Parameter( - torch.Tensor(self.shared_hid, self.shared_hid)) - self.w_hc = None - self.w_hh = None - - self.w_h = collections.defaultdict(dict) - self.w_c = collections.defaultdict(dict) - - for idx in range(self.num_blocks): - for jdx in range(idx + 1, self.num_blocks): - self.w_h[idx][jdx] = nn.Linear(self.shared_hid, - self.shared_hid, - bias=False) - self.w_c[idx][jdx] = nn.Linear(self.shared_hid, - self.shared_hid, - bias=False) - - self._w_h = nn.ModuleList([self.w_h[idx][jdx] - for idx in self.w_h - for jdx in self.w_h[idx]]) - self._w_c = nn.ModuleList([self.w_c[idx][jdx] - for idx in self.w_c - for jdx in self.w_c[idx]]) - - self.batch_norm = None - # if args.mode == 'train': - # self.batch_norm = nn.BatchNorm1d(self.shared_hid) - # else: - # self.batch_norm = None - - self.reset_parameters() - self.static_init_hidden = utils.keydefaultdict(self.init_hidden) - - def setDAG(self, dag): - if self.dag is None: - self.dag = dag - - def forward(self, word_seq, hidden=None): - inputs = torch.transpose(word_seq, 0, 1) - - time_steps = inputs.size(0) - batch_size = inputs.size(1) - - - self.w_hh = _get_dropped_weights(self.w_hh_raw, - 0.5, - self.training) - self.w_hc = _get_dropped_weights(self.w_hc_raw, - 0.5, - self.training) - - # hidden = self.static_init_hidden[batch_size] if hidden is None else hidden - hidden = self.static_init_hidden[batch_size] - - embed = self.encoder(inputs) - - embed = self.lockdrop(embed, 0.65 if self.training else 0) - - # The norm of hidden states are clipped here because - # otherwise ENAS is especially prone to exploding activations on the - # forward pass. This could probably be fixed in a more elegant way, but - # it might be exposing a weakness in the ENAS algorithm as currently - # proposed. - # - # For more details, see - # https://github.com/carpedm20/ENAS-pytorch/issues/6 - clipped_num = 0 - max_clipped_norm = 0 - h1tohT = [] - logits = [] - for step in range(time_steps): - x_t = embed[step] - logit, hidden = self.cell(x_t, hidden, self.dag) - - hidden_norms = hidden.norm(dim=-1) - max_norm = 25.0 - if hidden_norms.data.max() > max_norm: - # Just directly use the torch slice operations - # in PyTorch v0.4. - # - # This workaround for PyTorch v0.3.1 does everything in numpy, - # because the PyTorch slicing and slice assignment is too - # flaky. - hidden_norms = hidden_norms.data.cpu().numpy() - - clipped_num += 1 - if hidden_norms.max() > max_clipped_norm: - max_clipped_norm = hidden_norms.max() - - clip_select = hidden_norms > max_norm - clip_norms = hidden_norms[clip_select] - - mask = np.ones(hidden.size()) - normalizer = max_norm/clip_norms - normalizer = normalizer[:, np.newaxis] - - mask[clip_select] = normalizer - - if self.use_cuda: - hidden *= torch.autograd.Variable( - torch.FloatTensor(mask).cuda(), requires_grad=False) - else: - hidden *= torch.autograd.Variable( - torch.FloatTensor(mask), requires_grad=False) - logits.append(logit) - h1tohT.append(hidden) - - h1tohT = torch.stack(h1tohT) - output = torch.stack(logits) - raw_output = output - - output = self.lockdrop(output, 0.4 if self.training else 0) - - #Pooling - output = torch.mean(output, 0) - - decoded = self.decoder(output) - - extra_out = {'dropped': decoded, - 'hiddens': h1tohT, - 'raw': raw_output} - return {'pred': decoded, 'hidden': hidden, 'extra_out': extra_out} - - def cell(self, x, h_prev, dag): - """Computes a single pass through the discovered RNN cell.""" - c = {} - h = {} - f = {} - - f[0] = self.get_f(dag[-1][0].name) - c[0] = torch.sigmoid(self.w_xc(x) + F.linear(h_prev, self.w_hc, None)) - h[0] = (c[0]*f[0](self.w_xh(x) + F.linear(h_prev, self.w_hh, None)) + - (1 - c[0])*h_prev) - - leaf_node_ids = [] - q = collections.deque() - q.append(0) - - # Computes connections from the parent nodes `node_id` - # to their child nodes `next_id` recursively, skipping leaf nodes. A - # leaf node is a node whose id == `self.num_blocks`. - # - # Connections between parent i and child j should be computed as - # h_j = c_j*f_{ij}{(W^h_{ij}*h_i)} + (1 - c_j)*h_i, - # where c_j = \sigmoid{(W^c_{ij}*h_i)} - # - # See Training details from Section 3.1 of the paper. - # - # The following algorithm does a breadth-first (since `q.popleft()` is - # used) search over the nodes and computes all the hidden states. - while True: - if len(q) == 0: - break - - node_id = q.popleft() - nodes = dag[node_id] - - for next_node in nodes: - next_id = next_node.id - if next_id == self.num_blocks: - leaf_node_ids.append(node_id) - assert len(nodes) == 1, ('parent of leaf node should have ' - 'only one child') - continue - - w_h = self.w_h[node_id][next_id] - w_c = self.w_c[node_id][next_id] - - f[next_id] = self.get_f(next_node.name) - c[next_id] = torch.sigmoid(w_c(h[node_id])) - h[next_id] = (c[next_id]*f[next_id](w_h(h[node_id])) + - (1 - c[next_id])*h[node_id]) - - q.append(next_id) - - # Instead of averaging loose ends, perhaps there should - # be a set of separate unshared weights for each "loose" connection - # between each node in a cell and the output. - # - # As it stands, all weights W^h_{ij} are doing double duty by - # connecting both from i to j, as well as from i to the output. - - # average all the loose ends - leaf_nodes = [h[node_id] for node_id in leaf_node_ids] - output = torch.mean(torch.stack(leaf_nodes, 2), -1) - - # stabilizing the Updates of omega - if self.batch_norm is not None: - output = self.batch_norm(output) - - return output, h[self.num_blocks - 1] - - def init_hidden(self, batch_size): - zeros = torch.zeros(batch_size, self.shared_hid) - return utils.get_variable(zeros, self.use_cuda, requires_grad=False) - - def get_f(self, name): - name = name.lower() - if name == 'relu': - f = torch.relu - elif name == 'tanh': - f = torch.tanh - elif name == 'identity': - f = lambda x: x - elif name == 'sigmoid': - f = torch.sigmoid - return f - - - @property - def num_parameters(self): - def size(p): - return np.prod(p.size()) - return sum([size(param) for param in self.parameters()]) - - - def reset_parameters(self): - init_range = 0.025 - # init_range = 0.025 if self.args.mode == 'train' else 0.04 - for param in self.parameters(): - param.data.uniform_(-init_range, init_range) - self.decoder.bias.data.fill_(0) - - def predict(self, word_seq): - """ - - :param word_seq: torch.LongTensor, [batch_size, seq_len] - :return predict: dict of torch.LongTensor, [batch_size, seq_len] - """ - output = self(word_seq) - _, predict = output['pred'].max(dim=1) - return {'pred': predict} diff --git a/legacy/automl/enas_trainer.py b/legacy/automl/enas_trainer.py deleted file mode 100644 index e3524aa9..00000000 --- a/legacy/automl/enas_trainer.py +++ /dev/null @@ -1,383 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch - -import math -import time -from datetime import datetime -from datetime import timedelta - -import numpy as np -import torch - -try: - from tqdm.auto import tqdm -except: - from fastNLP.core.utils import _pseudo_tqdm as tqdm - -from fastNLP.core.batch import Batch -from fastNLP.core.callback import CallbackException -from fastNLP.core.dataset import DataSet -from fastNLP.core.utils import _move_dict_value_to_device -import fastNLP -from . import enas_utils as utils -from fastNLP.core.utils import _build_args - -from torch.optim import Adam - - -def _get_no_grad_ctx_mgr(): - """Returns a the `torch.no_grad` context manager for PyTorch version >= - 0.4, or a no-op context manager otherwise. - """ - return torch.no_grad() - - -class ENASTrainer(fastNLP.Trainer): - """A class to wrap training code.""" - def __init__(self, train_data, model, controller, **kwargs): - """Constructor for training algorithm. - :param DataSet train_data: the training data - :param torch.nn.modules.module model: a PyTorch model - :param torch.nn.modules.module controller: a PyTorch model - """ - self.final_epochs = kwargs['final_epochs'] - kwargs.pop('final_epochs') - super(ENASTrainer, self).__init__(train_data, model, **kwargs) - self.controller_step = 0 - self.shared_step = 0 - self.max_length = 35 - - self.shared = model - self.controller = controller - - self.shared_optim = Adam( - self.shared.parameters(), - lr=20.0, - weight_decay=1e-7) - - self.controller_optim = Adam( - self.controller.parameters(), - lr=3.5e-4) - - def train(self, load_best_model=True): - """ - :param bool load_best_model: 该参数只有在初始化提供了dev_data的情况下有效，如果True, trainer将在返回之前重新加载dev表现 - 最好的模型参数。 - :return results: 返回一个字典类型的数据, - 内含以下内容:: - - seconds: float, 表示训练时长 - 以下三个内容只有在提供了dev_data的情况下会有。 - best_eval: Dict of Dict, 表示evaluation的结果 - best_epoch: int，在第几个epoch取得的最佳值 - best_step: int, 在第几个step(batch)更新取得的最佳值 - - """ - results = {} - if self.n_epochs <= 0: - print(f"training epoch is {self.n_epochs}, nothing was done.") - results['seconds'] = 0. - return results - try: - if torch.cuda.is_available() and self.use_cuda: - self.model = self.model.cuda() - self._model_device = self.model.parameters().__next__().device - self._mode(self.model, is_test=False) - - self.start_time = str(datetime.now().strftime('%Y-%m-%d-%H-%M-%S')) - start_time = time.time() - print("training epochs started " + self.start_time, flush=True) - - try: - self.callback_manager.on_train_begin() - self._train() - self.callback_manager.on_train_end(self.model) - except (CallbackException, KeyboardInterrupt) as e: - self.callback_manager.on_exception(e, self.model) - - if self.dev_data is not None: - print("\nIn Epoch:{}/Step:{}, got best dev performance:".format(self.best_dev_epoch, self.best_dev_step) + - self.tester._format_eval_results(self.best_dev_perf),) - results['best_eval'] = self.best_dev_perf - results['best_epoch'] = self.best_dev_epoch - results['best_step'] = self.best_dev_step - if load_best_model: - model_name = "best_" + "_".join([self.model.__class__.__name__, self.metric_key, self.start_time]) - load_succeed = self._load_model(self.model, model_name) - if load_succeed: - print("Reloaded the best model.") - else: - print("Fail to reload best model.") - finally: - pass - results['seconds'] = round(time.time() - start_time, 2) - - return results - - def _train(self): - if not self.use_tqdm: - from fastNLP.core.utils import _pseudo_tqdm as inner_tqdm - else: - inner_tqdm = tqdm - self.step = 0 - start = time.time() - total_steps = (len(self.train_data) // self.batch_size + int( - len(self.train_data) % self.batch_size != 0)) * self.n_epochs - with inner_tqdm(total=total_steps, postfix='loss:{0:<6.5f}', leave=False, dynamic_ncols=True) as pbar: - avg_loss = 0 - data_iterator = Batch(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) - for epoch in range(1, self.n_epochs+1): - pbar.set_description_str(desc="Epoch {}/{}".format(epoch, self.n_epochs)) - last_stage = (epoch > self.n_epochs + 1 - self.final_epochs) - if epoch == self.n_epochs + 1 - self.final_epochs: - print('Entering the final stage. (Only train the selected structure)') - # early stopping - self.callback_manager.on_epoch_begin(epoch, self.n_epochs) - - # 1. Training the shared parameters omega of the child models - self.train_shared(pbar) - - # 2. Training the controller parameters theta - if not last_stage: - self.train_controller() - - if ((self.validate_every > 0 and self.step % self.validate_every == 0) or - (self.validate_every < 0 and self.step % len(data_iterator) == 0)) \ - and self.dev_data is not None: - if not last_stage: - self.derive() - eval_res = self._do_validation(epoch=epoch, step=self.step) - eval_str = "Evaluation at Epoch {}/{}. Step:{}/{}. ".format(epoch, self.n_epochs, self.step, - total_steps) + \ - self.tester._format_eval_results(eval_res) - pbar.write(eval_str) - - # lr decay; early stopping - self.callback_manager.on_epoch_end(epoch, self.n_epochs, self.optimizer) - # =============== epochs end =================== # - pbar.close() - # ============ tqdm end ============== # - - - def get_loss(self, inputs, targets, hidden, dags): - """Computes the loss for the same batch for M models. - - This amounts to an estimate of the loss, which is turned into an - estimate for the gradients of the shared model. - """ - if not isinstance(dags, list): - dags = [dags] - - loss = 0 - for dag in dags: - self.shared.setDAG(dag) - inputs = _build_args(self.shared.forward, **inputs) - inputs['hidden'] = hidden - result = self.shared(**inputs) - output, hidden, extra_out = result['pred'], result['hidden'], result['extra_out'] - - self.callback_manager.on_loss_begin(targets, result) - sample_loss = self._compute_loss(result, targets) - loss += sample_loss - - assert len(dags) == 1, 'there are multiple `hidden` for multple `dags`' - return loss, hidden, extra_out - - def train_shared(self, pbar=None, max_step=None, dag=None): - """Train the language model for 400 steps of minibatches of 64 - examples. - - Args: - max_step: Used to run extra training steps as a warm-up. - dag: If not None, is used instead of calling sample(). - - BPTT is truncated at 35 timesteps. - - For each weight update, gradients are estimated by sampling M models - from the fixed controller policy, and averaging their gradients - computed on a batch of training data. - """ - model = self.shared - model.train() - self.controller.eval() - - hidden = self.shared.init_hidden(self.batch_size) - - abs_max_grad = 0 - abs_max_hidden_norm = 0 - step = 0 - raw_total_loss = 0 - total_loss = 0 - train_idx = 0 - avg_loss = 0 - data_iterator = Batch(self.train_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) - - for batch_x, batch_y in data_iterator: - _move_dict_value_to_device(batch_x, batch_y, device=self._model_device) - indices = data_iterator.get_batch_indices() - # negative sampling; replace unknown; re-weight batch_y - self.callback_manager.on_batch_begin(batch_x, batch_y, indices) - # prediction = self._data_forward(self.model, batch_x) - - dags = self.controller.sample(1) - inputs, targets = batch_x, batch_y - # self.callback_manager.on_loss_begin(batch_y, prediction) - loss, hidden, extra_out = self.get_loss(inputs, - targets, - hidden, - dags) - hidden.detach_() - - avg_loss += loss.item() - - # Is loss NaN or inf? requires_grad = False - self.callback_manager.on_backward_begin(loss, self.model) - self._grad_backward(loss) - self.callback_manager.on_backward_end(self.model) - - self._update() - self.callback_manager.on_step_end(self.optimizer) - - if (self.step+1) % self.print_every == 0: - if self.use_tqdm: - print_output = "loss:{0:<6.5f}".format(avg_loss / self.print_every) - pbar.update(self.print_every) - else: - end = time.time() - diff = timedelta(seconds=round(end - start)) - print_output = "[epoch: {:>3} step: {:>4}] train loss: {:>4.6} time: {}".format( - epoch, self.step, avg_loss, diff) - pbar.set_postfix_str(print_output) - avg_loss = 0 - self.step += 1 - step += 1 - self.shared_step += 1 - self.callback_manager.on_batch_end() - # ================= mini-batch end ==================== # - - - def get_reward(self, dag, entropies, hidden, valid_idx=0): - """Computes the perplexity of a single sampled model on a minibatch of - validation data. - """ - if not isinstance(entropies, np.ndarray): - entropies = entropies.data.cpu().numpy() - - data_iterator = Batch(self.dev_data, batch_size=self.batch_size, sampler=self.sampler, as_numpy=False, - prefetch=self.prefetch) - - for inputs, targets in data_iterator: - valid_loss, hidden, _ = self.get_loss(inputs, targets, hidden, dag) - valid_loss = utils.to_item(valid_loss.data) - - valid_ppl = math.exp(valid_loss) - - R = 80 / valid_ppl - - rewards = R + 1e-4 * entropies - - return rewards, hidden - - def train_controller(self): - """Fixes the shared parameters and updates the controller parameters. - - The controller is updated with a score function gradient estimator - (i.e., REINFORCE), with the reward being c/valid_ppl, where valid_ppl - is computed on a minibatch of validation data. - - A moving average baseline is used. - - The controller is trained for 2000 steps per epoch (i.e., - first (Train Shared) phase -> second (Train Controller) phase). - """ - model = self.controller - model.train() - # Why can't we call shared.eval() here? Leads to loss - # being uniformly zero for the controller. - # self.shared.eval() - - avg_reward_base = None - baseline = None - adv_history = [] - entropy_history = [] - reward_history = [] - - hidden = self.shared.init_hidden(self.batch_size) - total_loss = 0 - valid_idx = 0 - for step in range(20): - # sample models - dags, log_probs, entropies = self.controller.sample( - with_details=True) - - # calculate reward - np_entropies = entropies.data.cpu().numpy() - # No gradients should be backpropagated to the - # shared model during controller training, obviously. - with _get_no_grad_ctx_mgr(): - rewards, hidden = self.get_reward(dags, - np_entropies, - hidden, - valid_idx) - - - reward_history.extend(rewards) - entropy_history.extend(np_entropies) - - # moving average baseline - if baseline is None: - baseline = rewards - else: - decay = 0.95 - baseline = decay * baseline + (1 - decay) * rewards - - adv = rewards - baseline - adv_history.extend(adv) - - # policy loss - loss = -log_probs*utils.get_variable(adv, - self.use_cuda, - requires_grad=False) - - loss = loss.sum() # or loss.mean() - - # update - self.controller_optim.zero_grad() - loss.backward() - - self.controller_optim.step() - - total_loss += utils.to_item(loss.data) - - if ((step % 50) == 0) and (step > 0): - reward_history, adv_history, entropy_history = [], [], [] - total_loss = 0 - - self.controller_step += 1 - # prev_valid_idx = valid_idx - # valid_idx = ((valid_idx + self.max_length) % - # (self.valid_data.size(0) - 1)) - # # Whenever we wrap around to the beginning of the - # # validation data, we reset the hidden states. - # if prev_valid_idx > valid_idx: - # hidden = self.shared.init_hidden(self.batch_size) - - def derive(self, sample_num=10, valid_idx=0): - """We are always deriving based on the very first batch - of validation data? This seems wrong... - """ - hidden = self.shared.init_hidden(self.batch_size) - - dags, _, entropies = self.controller.sample(sample_num, - with_details=True) - - max_R = 0 - best_dag = None - for dag in dags: - R, _ = self.get_reward(dag, entropies, hidden, valid_idx) - if R.max() > max_R: - max_R = R.max() - best_dag = dag - - self.model.setDAG(best_dag) diff --git a/legacy/automl/enas_utils.py b/legacy/automl/enas_utils.py deleted file mode 100644 index 7a53dd12..00000000 --- a/legacy/automl/enas_utils.py +++ /dev/null @@ -1,53 +0,0 @@ -# Code Modified from https://github.com/carpedm20/ENAS-pytorch - -from __future__ import print_function - -import collections -from collections import defaultdict - -import numpy as np -import torch -from torch.autograd import Variable - - -def detach(h): - if type(h) == Variable: - return Variable(h.data) - else: - return tuple(detach(v) for v in h) - -def get_variable(inputs, cuda=False, **kwargs): - if type(inputs) in [list, np.ndarray]: - inputs = torch.Tensor(inputs) - if cuda: - out = Variable(inputs.cuda(), **kwargs) - else: - out = Variable(inputs, **kwargs) - return out - -def update_lr(optimizer, lr): - for param_group in optimizer.param_groups: - param_group['lr'] = lr - -Node = collections.namedtuple('Node', ['id', 'name']) - - -class keydefaultdict(defaultdict): - def __missing__(self, key): - if self.default_factory is None: - raise KeyError(key) - else: - ret = self[key] = self.default_factory(key) - return ret - - -def to_item(x): - """Converts x, possibly scalar and possibly tensor, to a Python scalar.""" - if isinstance(x, (float, int)): - return x - - if float(torch.__version__[0:3]) < 0.4: - assert (x.dim() == 1) and (len(x) == 1) - return x[0] - - return x.item() diff --git a/legacy/component/__init__.py b/legacy/component/__init__.py deleted file mode 100644 index c6784aef..00000000 --- a/legacy/component/__init__.py +++ /dev/null @@ -1 +0,0 @@ -from .bert_tokenizer import BertTokenizer diff --git a/legacy/component/bert_tokenizer.py b/legacy/component/bert_tokenizer.py deleted file mode 100644 index 6354076d..00000000 --- a/legacy/component/bert_tokenizer.py +++ /dev/null @@ -1,378 +0,0 @@ -""" -bert_tokenizer.py is modified from huggingface/pytorch-pretrained-BERT, which is licensed under the Apache License 2.0. -""" -import collections -import os -import unicodedata -from io import open - - -PRETRAINED_VOCAB_ARCHIVE_MAP = { - 'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-vocab.txt", - 'bert-large-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-vocab.txt", - 'bert-base-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-vocab.txt", - 'bert-large-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-cased-vocab.txt", - 'bert-base-multilingual-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-uncased-vocab.txt", - 'bert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-vocab.txt", - 'bert-base-chinese': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-vocab.txt", -} -PRETRAINED_VOCAB_POSITIONAL_EMBEDDINGS_SIZE_MAP = { - 'bert-base-uncased': 512, - 'bert-large-uncased': 512, - 'bert-base-cased': 512, - 'bert-large-cased': 512, - 'bert-base-multilingual-uncased': 512, - 'bert-base-multilingual-cased': 512, - 'bert-base-chinese': 512, -} -VOCAB_NAME = 'vocab.txt' - - -def load_vocab(vocab_file): - """Loads a vocabulary file into a dictionary.""" - vocab = collections.OrderedDict() - index = 0 - with open(vocab_file, "r", encoding="utf-8") as reader: - while True: - token = reader.readline() - if not token: - break - token = token.strip() - vocab[token] = index - index += 1 - return vocab - - -def whitespace_tokenize(text): - """Runs basic whitespace cleaning and splitting on a piece of text.""" - text = text.strip() - if not text: - return [] - tokens = text.split() - return tokens - - -class BertTokenizer(object): - """Runs end-to-end tokenization: punctuation splitting + wordpiece""" - - def __init__(self, vocab_file, do_lower_case=True, max_len=None, do_basic_tokenize=True, - never_split=("[UNK]", "[SEP]", "[PAD]", "[CLS]", "[MASK]")): - """Constructs a BertTokenizer. - Args: - vocab_file: Path to a one-wordpiece-per-line vocabulary file - do_lower_case: Whether to lower case the input - Only has an effect when do_wordpiece_only=False - do_basic_tokenize: Whether to do basic tokenization before wordpiece. - max_len: An artificial maximum length to truncate tokenized sequences to; - Effective maximum length is always the minimum of this - value (if specified) and the underlying BERT model's - sequence length. - never_split: List of tokens which will never be split during tokenization. - Only has an effect when do_wordpiece_only=False - """ - if not os.path.isfile(vocab_file): - raise ValueError( - "Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained " - "model use `tokenizer = BertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file)) - self.vocab = load_vocab(vocab_file) - self.ids_to_tokens = collections.OrderedDict( - [(ids, tok) for tok, ids in self.vocab.items()]) - self.do_basic_tokenize = do_basic_tokenize - if do_basic_tokenize: - self.basic_tokenizer = BasicTokenizer(do_lower_case=do_lower_case, - never_split=never_split) - self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab) - self.max_len = max_len if max_len is not None else int(1e12) - - def tokenize(self, text): - split_tokens = [] - if self.do_basic_tokenize: - for token in self.basic_tokenizer.tokenize(text): - for sub_token in self.wordpiece_tokenizer.tokenize(token): - split_tokens.append(sub_token) - else: - split_tokens = self.wordpiece_tokenizer.tokenize(text) - return split_tokens - - def convert_tokens_to_ids(self, tokens): - """Converts a sequence of tokens into ids using the vocab.""" - ids = [] - for token in tokens: - ids.append(self.vocab[token]) - if len(ids) > self.max_len: - print( - "WARNING!\n\"" - "Token indices sequence length is longer than the specified maximum " - "sequence length for this BERT model ({} > {}). Running this" - " sequence through BERT will result in indexing errors".format(len(ids), self.max_len) - ) - return ids - - def convert_ids_to_tokens(self, ids): - """Converts a sequence of ids in wordpiece tokens using the vocab.""" - tokens = [] - for i in ids: - tokens.append(self.ids_to_tokens[i]) - return tokens - - def save_vocabulary(self, vocab_path): - """Save the tokenizer vocabulary to a directory or file.""" - index = 0 - if os.path.isdir(vocab_path): - vocab_file = os.path.join(vocab_path, VOCAB_NAME) - with open(vocab_file, "w", encoding="utf-8") as writer: - for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]): - if index != token_index: - print("Saving vocabulary to {}: vocabulary indices are not consecutive." - " Please check that the vocabulary is not corrupted!".format(vocab_file)) - index = token_index - writer.write(token + u'\n') - index += 1 - return vocab_file - - @classmethod - def from_pretrained(cls, pretrained_model_name_or_path, cache_dir=None, *inputs, **kwargs): - """ - Instantiate a PreTrainedBertModel from a pre-trained model file. - Download and cache the pre-trained model file if needed. - """ - if pretrained_model_name_or_path in PRETRAINED_VOCAB_ARCHIVE_MAP: - vocab_file = PRETRAINED_VOCAB_ARCHIVE_MAP[pretrained_model_name_or_path] - if '-cased' in pretrained_model_name_or_path and kwargs.get('do_lower_case', True): - print("The pre-trained model you are loading is a cased model but you have not set " - "`do_lower_case` to False. We are setting `do_lower_case=False` for you but " - "you may want to check this behavior.") - kwargs['do_lower_case'] = False - elif '-cased' not in pretrained_model_name_or_path and not kwargs.get('do_lower_case', True): - print("The pre-trained model you are loading is an uncased model but you have set " - "`do_lower_case` to False. We are setting `do_lower_case=True` for you " - "but you may want to check this behavior.") - kwargs['do_lower_case'] = True - else: - vocab_file = pretrained_model_name_or_path - if os.path.isdir(vocab_file): - vocab_file = os.path.join(vocab_file, VOCAB_NAME) - # redirect to the cache, if necessary - resolved_vocab_file = vocab_file - print("loading vocabulary file {}".format(vocab_file)) - if pretrained_model_name_or_path in PRETRAINED_VOCAB_POSITIONAL_EMBEDDINGS_SIZE_MAP: - # if we're using a pretrained model, ensure the tokenizer wont index sequences longer - # than the number of positional embeddings - max_len = PRETRAINED_VOCAB_POSITIONAL_EMBEDDINGS_SIZE_MAP[pretrained_model_name_or_path] - kwargs['max_len'] = min(kwargs.get('max_len', int(1e12)), max_len) - # Instantiate tokenizer. - tokenizer = cls(resolved_vocab_file, *inputs, **kwargs) - return tokenizer - - -class BasicTokenizer(object): - """Runs basic tokenization (punctuation splitting, lower casing, etc.).""" - - def __init__(self, - do_lower_case=True, - never_split=("[UNK]", "[SEP]", "[PAD]", "[CLS]", "[MASK]")): - """Constructs a BasicTokenizer. - Args: - do_lower_case: Whether to lower case the input. - """ - self.do_lower_case = do_lower_case - self.never_split = never_split - - def tokenize(self, text): - """Tokenizes a piece of text.""" - text = self._clean_text(text) - # This was added on November 1st, 2018 for the multilingual and Chinese - # models. This is also applied to the English models now, but it doesn't - # matter since the English models were not trained on any Chinese data - # and generally don't have any Chinese data in them (there are Chinese - # characters in the vocabulary because Wikipedia does have some Chinese - # words in the English Wikipedia.). - text = self._tokenize_chinese_chars(text) - orig_tokens = whitespace_tokenize(text) - split_tokens = [] - for token in orig_tokens: - if self.do_lower_case and token not in self.never_split: - token = token.lower() - token = self._run_strip_accents(token) - split_tokens.extend(self._run_split_on_punc(token)) - - output_tokens = whitespace_tokenize(" ".join(split_tokens)) - return output_tokens - - def _run_strip_accents(self, text): - """Strips accents from a piece of text.""" - text = unicodedata.normalize("NFD", text) - output = [] - for char in text: - cat = unicodedata.category(char) - if cat == "Mn": - continue - output.append(char) - return "".join(output) - - def _run_split_on_punc(self, text): - """Splits punctuation on a piece of text.""" - if text in self.never_split: - return [text] - chars = list(text) - i = 0 - start_new_word = True - output = [] - while i < len(chars): - char = chars[i] - if _is_punctuation(char): - output.append([char]) - start_new_word = True - else: - if start_new_word: - output.append([]) - start_new_word = False - output[-1].append(char) - i += 1 - - return ["".join(x) for x in output] - - def _tokenize_chinese_chars(self, text): - """Adds whitespace around any CJK character.""" - output = [] - for char in text: - cp = ord(char) - if self._is_chinese_char(cp): - output.append(" ") - output.append(char) - output.append(" ") - else: - output.append(char) - return "".join(output) - - def _is_chinese_char(self, cp): - """Checks whether CP is the codepoint of a CJK character.""" - # This defines a "chinese character" as anything in the CJK Unicode block: - # https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block) - # - # Note that the CJK Unicode block is NOT all Japanese and Korean characters, - # despite its name. The modern Korean Hangul alphabet is a different block, - # as is Japanese Hiragana and Katakana. Those alphabets are used to write - # space-separated words, so they are not treated specially and handled - # like the all of the other languages. - if ((cp >= 0x4E00 and cp <= 0x9FFF) or # - (cp >= 0x3400 and cp <= 0x4DBF) or # - (cp >= 0x20000 and cp <= 0x2A6DF) or # - (cp >= 0x2A700 and cp <= 0x2B73F) or # - (cp >= 0x2B740 and cp <= 0x2B81F) or # - (cp >= 0x2B820 and cp <= 0x2CEAF) or - (cp >= 0xF900 and cp <= 0xFAFF) or # - (cp >= 0x2F800 and cp <= 0x2FA1F)): # - return True - - return False - - def _clean_text(self, text): - """Performs invalid character removal and whitespace cleanup on text.""" - output = [] - for char in text: - cp = ord(char) - if cp == 0 or cp == 0xfffd or _is_control(char): - continue - if _is_whitespace(char): - output.append(" ") - else: - output.append(char) - return "".join(output) - - -class WordpieceTokenizer(object): - """Runs WordPiece tokenization.""" - - def __init__(self, vocab, unk_token="[UNK]", max_input_chars_per_word=100): - self.vocab = vocab - self.unk_token = unk_token - self.max_input_chars_per_word = max_input_chars_per_word - - def tokenize(self, text): - """Tokenizes a piece of text into its word pieces. - This uses a greedy longest-match-first algorithm to perform tokenization - using the given vocabulary. - For example: - input = "unaffable" - output = ["un", "##aff", "##able"] - Args: - text: A single token or whitespace separated tokens. This should have - already been passed through `BasicTokenizer`. - Returns: - A list of wordpiece tokens. - """ - - output_tokens = [] - for token in whitespace_tokenize(text): - chars = list(token) - if len(chars) > self.max_input_chars_per_word: - output_tokens.append(self.unk_token) - continue - - is_bad = False - start = 0 - sub_tokens = [] - while start < len(chars): - end = len(chars) - cur_substr = None - while start < end: - substr = "".join(chars[start:end]) - if start > 0: - substr = "##" + substr - if substr in self.vocab: - cur_substr = substr - break - end -= 1 - if cur_substr is None: - is_bad = True - break - sub_tokens.append(cur_substr) - start = end - - if is_bad: - output_tokens.append(self.unk_token) - else: - output_tokens.extend(sub_tokens) - return output_tokens - - -def _is_whitespace(char): - """Checks whether `chars` is a whitespace character.""" - # \t, \n, and \r are technically contorl characters but we treat them - # as whitespace since they are generally considered as such. - if char == " " or char == "\t" or char == "\n" or char == "\r": - return True - cat = unicodedata.category(char) - if cat == "Zs": - return True - return False - - -def _is_control(char): - """Checks whether `chars` is a control character.""" - # These are technically control characters but we count them as whitespace - # characters. - if char == "\t" or char == "\n" or char == "\r": - return False - cat = unicodedata.category(char) - if cat.startswith("C"): - return True - return False - - -def _is_punctuation(char): - """Checks whether `chars` is a punctuation character.""" - cp = ord(char) - # We treat all non-letter/number ASCII as punctuation. - # Characters such as "^", "$", and "`" are not in the Unicode - # Punctuation class but we treat them as punctuation anyways, for - # consistency. - if ((cp >= 33 and cp <= 47) or (cp >= 58 and cp <= 64) or - (cp >= 91 and cp <= 96) or (cp >= 123 and cp <= 126)): - return True - cat = unicodedata.category(char) - if cat.startswith("P"): - return True - return False - diff --git a/reproduction/Summarization/Baseline/data/dataloader.py b/reproduction/Summarization/Baseline/data/dataloader.py index 47cd0856..dcb294b0 100644 --- a/reproduction/Summarization/Baseline/data/dataloader.py +++ b/reproduction/Summarization/Baseline/data/dataloader.py @@ -1,188 +1,188 @@ -import pickle -import numpy as np - -from fastNLP.core.vocabulary import Vocabulary -from fastNLP.io.base_loader import DataBundle -from fastNLP.io.dataset_loader import JsonLoader -from fastNLP.core.const import Const - -from tools.logger import * - -WORD_PAD = "[PAD]" -WORD_UNK = "[UNK]" -DOMAIN_UNK = "X" -TAG_UNK = "X" - - -class SummarizationLoader(JsonLoader): - """ - 读取summarization数据集，读取的DataSet包含fields:: - - text: list(str)，document - summary: list(str), summary - text_wd: list(list(str))，tokenized document - summary_wd: list(list(str)), tokenized summary - labels: list(int), - flatten_label: list(int), 0 or 1, flatten labels - domain: str, optional - tag: list(str), optional - - 数据来源: CNN_DailyMail Newsroom DUC - """ - - def __init__(self): - super(SummarizationLoader, self).__init__() - - def _load(self, path): - ds = super(SummarizationLoader, self)._load(path) - - def _lower_text(text_list): - return [text.lower() for text in text_list] - - def _split_list(text_list): - return [text.split() for text in text_list] - - def _convert_label(label, sent_len): - np_label = np.zeros(sent_len, dtype=int) - if label != []: - np_label[np.array(label)] = 1 - return np_label.tolist() - - ds.apply(lambda x: _lower_text(x['text']), new_field_name='text') - ds.apply(lambda x: _lower_text(x['summary']), new_field_name='summary') - ds.apply(lambda x:_split_list(x['text']), new_field_name='text_wd') - ds.apply(lambda x:_split_list(x['summary']), new_field_name='summary_wd') - ds.apply(lambda x:_convert_label(x["label"], len(x["text"])), new_field_name="flatten_label") - - return ds - - def process(self, paths, vocab_size, vocab_path, sent_max_len, doc_max_timesteps, domain=False, tag=False, load_vocab_file=True): - """ - :param paths: dict path for each dataset - :param vocab_size: int max_size for vocab - :param vocab_path: str vocab path - :param sent_max_len: int max token number of the sentence - :param doc_max_timesteps: int max sentence number of the document - :param domain: bool build vocab for publication, use 'X' for unknown - :param tag: bool build vocab for tag, use 'X' for unknown - :param load_vocab_file: bool build vocab (False) or load vocab (True) - :return: DataBundle - datasets: dict keys correspond to the paths dict - vocabs: dict key: vocab(if "train" in paths), domain(if domain=True), tag(if tag=True) - embeddings: optional - """ - - def _pad_sent(text_wd): - pad_text_wd = [] - for sent_wd in text_wd: - if len(sent_wd) < sent_max_len: - pad_num = sent_max_len - len(sent_wd) - sent_wd.extend([WORD_PAD] * pad_num) - else: - sent_wd = sent_wd[:sent_max_len] - pad_text_wd.append(sent_wd) - return pad_text_wd - - def _token_mask(text_wd): - token_mask_list = [] - for sent_wd in text_wd: - token_num = len(sent_wd) - if token_num < sent_max_len: - mask = [1] * token_num + [0] * (sent_max_len - token_num) - else: - mask = [1] * sent_max_len - token_mask_list.append(mask) - return token_mask_list - - def _pad_label(label): - text_len = len(label) - if text_len < doc_max_timesteps: - pad_label = label + [0] * (doc_max_timesteps - text_len) - else: - pad_label = label[:doc_max_timesteps] - return pad_label - - def _pad_doc(text_wd): - text_len = len(text_wd) - if text_len < doc_max_timesteps: - padding = [WORD_PAD] * sent_max_len - pad_text = text_wd + [padding] * (doc_max_timesteps - text_len) - else: - pad_text = text_wd[:doc_max_timesteps] - return pad_text - - def _sent_mask(text_wd): - text_len = len(text_wd) - if text_len < doc_max_timesteps: - sent_mask = [1] * text_len + [0] * (doc_max_timesteps - text_len) - else: - sent_mask = [1] * doc_max_timesteps - return sent_mask - - - datasets = {} - train_ds = None - for key, value in paths.items(): - ds = self.load(value) - # pad sent - ds.apply(lambda x:_pad_sent(x["text_wd"]), new_field_name="pad_text_wd") - ds.apply(lambda x:_token_mask(x["text_wd"]), new_field_name="pad_token_mask") - # pad document - ds.apply(lambda x:_pad_doc(x["pad_text_wd"]), new_field_name="pad_text") - ds.apply(lambda x:_sent_mask(x["pad_text_wd"]), new_field_name="seq_len") - ds.apply(lambda x:_pad_label(x["flatten_label"]), new_field_name="pad_label") - - # rename field - ds.rename_field("pad_text", Const.INPUT) - ds.rename_field("seq_len", Const.INPUT_LEN) - ds.rename_field("pad_label", Const.TARGET) - - # set input and target - ds.set_input(Const.INPUT, Const.INPUT_LEN) - ds.set_target(Const.TARGET, Const.INPUT_LEN) - - datasets[key] = ds - if "train" in key: - train_ds = datasets[key] - - vocab_dict = {} - if load_vocab_file == False: - logger.info("[INFO] Build new vocab from training dataset!") - if train_ds == None: - raise ValueError("Lack train file to build vocabulary!") - - vocabs = Vocabulary(max_size=vocab_size, padding=WORD_PAD, unknown=WORD_UNK) - vocabs.from_dataset(train_ds, field_name=["text_wd","summary_wd"]) - vocab_dict["vocab"] = vocabs - else: - logger.info("[INFO] Load existing vocab from %s!" % vocab_path) - word_list = [] - with open(vocab_path, 'r', encoding='utf8') as vocab_f: - cnt = 2 # pad and unk - for line in vocab_f: - pieces = line.split("\t") - word_list.append(pieces[0]) - cnt += 1 - if cnt > vocab_size: - break - vocabs = Vocabulary(max_size=vocab_size, padding=WORD_PAD, unknown=WORD_UNK) - vocabs.add_word_lst(word_list) - vocabs.build_vocab() - vocab_dict["vocab"] = vocabs - - if domain == True: - domaindict = Vocabulary(padding=None, unknown=DOMAIN_UNK) - domaindict.from_dataset(train_ds, field_name="publication") - vocab_dict["domain"] = domaindict - if tag == True: - tagdict = Vocabulary(padding=None, unknown=TAG_UNK) - tagdict.from_dataset(train_ds, field_name="tag") - vocab_dict["tag"] = tagdict - - for ds in datasets.values(): - vocab_dict["vocab"].index_dataset(ds, field_name=Const.INPUT, new_field_name=Const.INPUT) - - return DataBundle(vocabs=vocab_dict, datasets=datasets) - - - +import pickle +import numpy as np + +from fastNLP.core.vocabulary import Vocabulary +from fastNLP.io.data_bundle import DataBundle +from fastNLP.io.dataset_loader import JsonLoader +from fastNLP.core.const import Const + +from tools.logger import * + +WORD_PAD = "[PAD]" +WORD_UNK = "[UNK]" +DOMAIN_UNK = "X" +TAG_UNK = "X" + + +class SummarizationLoader(JsonLoader): + """ + 读取summarization数据集，读取的DataSet包含fields:: + + text: list(str)，document + summary: list(str), summary + text_wd: list(list(str))，tokenized document + summary_wd: list(list(str)), tokenized summary + labels: list(int), + flatten_label: list(int), 0 or 1, flatten labels + domain: str, optional + tag: list(str), optional + + 数据来源: CNN_DailyMail Newsroom DUC + """ + + def __init__(self): + super(SummarizationLoader, self).__init__() + + def _load(self, path): + ds = super(SummarizationLoader, self)._load(path) + + def _lower_text(text_list): + return [text.lower() for text in text_list] + + def _split_list(text_list): + return [text.split() for text in text_list] + + def _convert_label(label, sent_len): + np_label = np.zeros(sent_len, dtype=int) + if label != []: + np_label[np.array(label)] = 1 + return np_label.tolist() + + ds.apply(lambda x: _lower_text(x['text']), new_field_name='text') + ds.apply(lambda x: _lower_text(x['summary']), new_field_name='summary') + ds.apply(lambda x:_split_list(x['text']), new_field_name='text_wd') + ds.apply(lambda x:_split_list(x['summary']), new_field_name='summary_wd') + ds.apply(lambda x:_convert_label(x["label"], len(x["text"])), new_field_name="flatten_label") + + return ds + + def process(self, paths, vocab_size, vocab_path, sent_max_len, doc_max_timesteps, domain=False, tag=False, load_vocab_file=True): + """ + :param paths: dict path for each dataset + :param vocab_size: int max_size for vocab + :param vocab_path: str vocab path + :param sent_max_len: int max token number of the sentence + :param doc_max_timesteps: int max sentence number of the document + :param domain: bool build vocab for publication, use 'X' for unknown + :param tag: bool build vocab for tag, use 'X' for unknown + :param load_vocab_file: bool build vocab (False) or load vocab (True) + :return: DataBundle + datasets: dict keys correspond to the paths dict + vocabs: dict key: vocab(if "train" in paths), domain(if domain=True), tag(if tag=True) + embeddings: optional + """ + + def _pad_sent(text_wd): + pad_text_wd = [] + for sent_wd in text_wd: + if len(sent_wd) < sent_max_len: + pad_num = sent_max_len - len(sent_wd) + sent_wd.extend([WORD_PAD] * pad_num) + else: + sent_wd = sent_wd[:sent_max_len] + pad_text_wd.append(sent_wd) + return pad_text_wd + + def _token_mask(text_wd): + token_mask_list = [] + for sent_wd in text_wd: + token_num = len(sent_wd) + if token_num < sent_max_len: + mask = [1] * token_num + [0] * (sent_max_len - token_num) + else: + mask = [1] * sent_max_len + token_mask_list.append(mask) + return token_mask_list + + def _pad_label(label): + text_len = len(label) + if text_len < doc_max_timesteps: + pad_label = label + [0] * (doc_max_timesteps - text_len) + else: + pad_label = label[:doc_max_timesteps] + return pad_label + + def _pad_doc(text_wd): + text_len = len(text_wd) + if text_len < doc_max_timesteps: + padding = [WORD_PAD] * sent_max_len + pad_text = text_wd + [padding] * (doc_max_timesteps - text_len) + else: + pad_text = text_wd[:doc_max_timesteps] + return pad_text + + def _sent_mask(text_wd): + text_len = len(text_wd) + if text_len < doc_max_timesteps: + sent_mask = [1] * text_len + [0] * (doc_max_timesteps - text_len) + else: + sent_mask = [1] * doc_max_timesteps + return sent_mask + + + datasets = {} + train_ds = None + for key, value in paths.items(): + ds = self.load(value) + # pad sent + ds.apply(lambda x:_pad_sent(x["text_wd"]), new_field_name="pad_text_wd") + ds.apply(lambda x:_token_mask(x["text_wd"]), new_field_name="pad_token_mask") + # pad document + ds.apply(lambda x:_pad_doc(x["pad_text_wd"]), new_field_name="pad_text") + ds.apply(lambda x:_sent_mask(x["pad_text_wd"]), new_field_name="seq_len") + ds.apply(lambda x:_pad_label(x["flatten_label"]), new_field_name="pad_label") + + # rename field + ds.rename_field("pad_text", Const.INPUT) + ds.rename_field("seq_len", Const.INPUT_LEN) + ds.rename_field("pad_label", Const.TARGET) + + # set input and target + ds.set_input(Const.INPUT, Const.INPUT_LEN) + ds.set_target(Const.TARGET, Const.INPUT_LEN) + + datasets[key] = ds + if "train" in key: + train_ds = datasets[key] + + vocab_dict = {} + if load_vocab_file == False: + logger.info("[INFO] Build new vocab from training dataset!") + if train_ds == None: + raise ValueError("Lack train file to build vocabulary!") + + vocabs = Vocabulary(max_size=vocab_size, padding=WORD_PAD, unknown=WORD_UNK) + vocabs.from_dataset(train_ds, field_name=["text_wd","summary_wd"]) + vocab_dict["vocab"] = vocabs + else: + logger.info("[INFO] Load existing vocab from %s!" % vocab_path) + word_list = [] + with open(vocab_path, 'r', encoding='utf8') as vocab_f: + cnt = 2 # pad and unk + for line in vocab_f: + pieces = line.split("\t") + word_list.append(pieces[0]) + cnt += 1 + if cnt > vocab_size: + break + vocabs = Vocabulary(max_size=vocab_size, padding=WORD_PAD, unknown=WORD_UNK) + vocabs.add_word_lst(word_list) + vocabs.build_vocab() + vocab_dict["vocab"] = vocabs + + if domain == True: + domaindict = Vocabulary(padding=None, unknown=DOMAIN_UNK) + domaindict.from_dataset(train_ds, field_name="publication") + vocab_dict["domain"] = domaindict + if tag == True: + tagdict = Vocabulary(padding=None, unknown=TAG_UNK) + tagdict.from_dataset(train_ds, field_name="tag") + vocab_dict["tag"] = tagdict + + for ds in datasets.values(): + vocab_dict["vocab"].index_dataset(ds, field_name=Const.INPUT, new_field_name=Const.INPUT) + + return DataBundle(vocabs=vocab_dict, datasets=datasets) + + + diff --git a/reproduction/Summarization/BertSum/dataloader.py b/reproduction/Summarization/BertSum/dataloader.py index c5201261..6af797e4 100644 --- a/reproduction/Summarization/BertSum/dataloader.py +++ b/reproduction/Summarization/BertSum/dataloader.py @@ -3,7 +3,7 @@ from datetime import timedelta from fastNLP.io.dataset_loader import JsonLoader from fastNLP.modules.encoder._bert import BertTokenizer -from fastNLP.io.base_loader import DataBundle +from fastNLP.io.data_bundle import DataBundle from fastNLP.core.const import Const class BertData(JsonLoader): diff --git a/reproduction/coreference_resolution/data_load/cr_loader.py b/reproduction/coreference_resolution/data_load/cr_loader.py index a424b0d1..5ed73473 100644 --- a/reproduction/coreference_resolution/data_load/cr_loader.py +++ b/reproduction/coreference_resolution/data_load/cr_loader.py @@ -1,7 +1,7 @@ from fastNLP.io.dataset_loader import JsonLoader,DataSet,Instance from fastNLP.io.file_reader import _read_json from fastNLP.core.vocabulary import Vocabulary -from fastNLP.io.base_loader import DataBundle +from fastNLP.io.data_bundle import DataBundle from reproduction.coreference_resolution.model.config import Config import reproduction.coreference_resolution.model.preprocess as preprocess diff --git a/reproduction/joint_cws_parse/data/data_loader.py b/reproduction/joint_cws_parse/data/data_loader.py index 3e6fec4b..4df46b04 100644 --- a/reproduction/joint_cws_parse/data/data_loader.py +++ b/reproduction/joint_cws_parse/data/data_loader.py @@ -1,6 +1,6 @@ -from fastNLP.io.base_loader import DataSetLoader, DataBundle +from fastNLP.io.data_bundle import DataSetLoader, DataBundle from fastNLP.io.data_loader import ConllLoader import numpy as np diff --git a/reproduction/joint_cws_parse/models/CharParser.py b/reproduction/joint_cws_parse/models/CharParser.py index c07c070e..7d89cacb 100644 --- a/reproduction/joint_cws_parse/models/CharParser.py +++ b/reproduction/joint_cws_parse/models/CharParser.py @@ -224,11 +224,11 @@ class CharBiaffineParser(BiaffineParser): batch_size, seq_len, _ = arc_pred.shape flip_mask = (mask == 0) - _arc_pred = arc_pred.clone() - _arc_pred.masked_fill_(flip_mask.unsqueeze(1), -float('inf')) + # _arc_pred = arc_pred.clone() + _arc_pred = arc_pred.masked_fill(flip_mask.unsqueeze(1), -float('inf')) - arc_true[:, 0].fill_(-1) - label_true[:, 0].fill_(-1) + arc_true.data[:, 0].fill_(-1) + label_true.data[:, 0].fill_(-1) arc_nll = F.cross_entropy(_arc_pred.view(-1, seq_len), arc_true.view(-1), ignore_index=-1) label_nll = F.cross_entropy(label_pred.view(-1, label_pred.size(-1)), label_true.view(-1), ignore_index=-1) diff --git a/reproduction/joint_cws_parse/train.py b/reproduction/joint_cws_parse/train.py index 0c34614b..ed4b07f0 100644 --- a/reproduction/joint_cws_parse/train.py +++ b/reproduction/joint_cws_parse/train.py @@ -14,6 +14,7 @@ from torch.optim.lr_scheduler import StepLR from fastNLP import Tester from fastNLP import GradientClipCallback, LRScheduler import os +from fastNLP import cache_results def set_random_seed(random_seed=666): import random, numpy, torch @@ -39,43 +40,42 @@ label_mlp_size = 100 batch_size = 32 update_every = 4 n_epochs = 100 -data_folder = '' # 填写在数据所在文件夹, 文件夹下应该有train, dev, test等三个文件 -vector_folder = '' # 预训练的vector，下面应该包含三个文件: 1grams_t3_m50_corpus.txt, 2grams_t3_m50_corpus.txt, 3grams_t3_m50_corpus.txt +data_name = 'new_ctb7' #################################################### +data_folder = f'/remote-home/hyan01/exps/JointCwsPosParser/data/{data_name}/output' # 填写在数据所在文件夹, 文件夹下应该有train, dev, test等三个文件 +vector_folder = '/remote-home/hyan01/exps/CWS/pretrain/vectors' # 预训练的vector，下面应该包含三个文件: 1grams_t3_m50_corpus.txt, 2grams_t3_m50_corpus.txt, 3grams_t3_m50_corpus.txt set_random_seed(1234) device = 0 -# @cache_results('caches/{}.pkl'.format(data_name)) -# def get_data(): -data = CTBxJointLoader().process(data_folder) - -char_labels_vocab = data.vocabs['char_labels'] - -pre_chars_vocab = data.vocabs['pre_chars'] -pre_bigrams_vocab = data.vocabs['pre_bigrams'] -pre_trigrams_vocab = data.vocabs['pre_trigrams'] - -chars_vocab = data.vocabs['chars'] -bigrams_vocab = data.vocabs['bigrams'] -trigrams_vocab = data.vocabs['trigrams'] - -pre_chars_embed = StaticEmbedding(pre_chars_vocab, - model_dir_or_name=os.path.join(vector_folder, '1grams_t3_m50_corpus.txt'), - init_method=uniform_init, normalize=False) -pre_chars_embed.embedding.weight.data = pre_chars_embed.embedding.weight.data/pre_chars_embed.embedding.weight.data.std() -pre_bigrams_embed = StaticEmbedding(pre_bigrams_vocab, - model_dir_or_name=os.path.join(vector_folder, '2grams_t3_m50_corpus.txt'), - init_method=uniform_init, normalize=False) -pre_bigrams_embed.embedding.weight.data = pre_bigrams_embed.embedding.weight.data/pre_bigrams_embed.embedding.weight.data.std() -pre_trigrams_embed = StaticEmbedding(pre_trigrams_vocab, - model_dir_or_name=os.path.join(vector_folder, '3grams_t3_m50_corpus.txt'), - init_method=uniform_init, normalize=False) -pre_trigrams_embed.embedding.weight.data = pre_trigrams_embed.embedding.weight.data/pre_trigrams_embed.embedding.weight.data.std() - - # return chars_vocab, bigrams_vocab, trigrams_vocab, char_labels_vocab, pre_chars_embed, pre_bigrams_embed, pre_trigrams_embed, data - -# chars_vocab, bigrams_vocab, trigrams_vocab, char_labels_vocab, pre_chars_embed, pre_bigrams_embed, pre_trigrams_embed, data = get_data() +@cache_results('caches/{}.pkl'.format(data_name)) +def get_data(): + data = CTBxJointLoader().process(data_folder) + char_labels_vocab = data.vocabs['char_labels'] + + pre_chars_vocab = data.vocabs['pre_chars'] + pre_bigrams_vocab = data.vocabs['pre_bigrams'] + pre_trigrams_vocab = data.vocabs['pre_trigrams'] + + chars_vocab = data.vocabs['chars'] + bigrams_vocab = data.vocabs['bigrams'] + trigrams_vocab = data.vocabs['trigrams'] + pre_chars_embed = StaticEmbedding(pre_chars_vocab, + model_dir_or_name=os.path.join(vector_folder, '1grams_t3_m50_corpus.txt'), + init_method=uniform_init, normalize=False) + pre_chars_embed.embedding.weight.data = pre_chars_embed.embedding.weight.data / pre_chars_embed.embedding.weight.data.std() + pre_bigrams_embed = StaticEmbedding(pre_bigrams_vocab, + model_dir_or_name=os.path.join(vector_folder, '2grams_t3_m50_corpus.txt'), + init_method=uniform_init, normalize=False) + pre_bigrams_embed.embedding.weight.data = pre_bigrams_embed.embedding.weight.data / pre_bigrams_embed.embedding.weight.data.std() + pre_trigrams_embed = StaticEmbedding(pre_trigrams_vocab, + model_dir_or_name=os.path.join(vector_folder, '3grams_t3_m50_corpus.txt'), + init_method=uniform_init, normalize=False) + pre_trigrams_embed.embedding.weight.data = pre_trigrams_embed.embedding.weight.data / pre_trigrams_embed.embedding.weight.data.std() + + return chars_vocab, bigrams_vocab, trigrams_vocab, char_labels_vocab, pre_chars_embed, pre_bigrams_embed, pre_trigrams_embed, data + +chars_vocab, bigrams_vocab, trigrams_vocab, char_labels_vocab, pre_chars_embed, pre_bigrams_embed, pre_trigrams_embed, data = get_data() print(data) model = CharParser(char_vocab_size=len(chars_vocab), @@ -104,11 +104,24 @@ optimizer = optim.Adam([param for param in model.parameters() if param.requires_ sampler = BucketSampler(seq_len_field_name='seq_lens') callbacks = [] + +from fastNLP.core.callback import Callback +from torch.optim.lr_scheduler import LambdaLR +class SchedulerCallback(Callback): + def __init__(self, scheduler): + super().__init__() + self.scheduler = scheduler + + def on_backward_end(self): + if self.step % self.update_every==0: + self.scheduler.step() + +scheduler = LambdaLR(optimizer, lr_lambda=lambda step:(0.75)**(step//5000)) # scheduler = LambdaLR(optimizer, lr_lambda=lambda step:(0.75)**(step//5000)) -scheduler = StepLR(optimizer, step_size=18, gamma=0.75) -# optim_callback = OptimizerCallback(optimizer, scheduler, update_every) +# scheduler = StepLR(optimizer, step_size=18, gamma=0.75) +scheduler_callback = SchedulerCallback(scheduler) # callbacks.append(optim_callback) -scheduler_callback = LRScheduler(scheduler) +# scheduler_callback = LRScheduler(scheduler) callbacks.append(scheduler_callback) callbacks.append(GradientClipCallback(clip_type='value', clip_value=5)) @@ -119,6 +132,6 @@ callbacks.append(dev_callback) trainer = Trainer(data.datasets['train'], model, loss=None, metrics=metrics, n_epochs=n_epochs, batch_size=batch_size, print_every=3, validate_every=-1, dev_data=data.datasets['dev'], save_path=None, optimizer=optimizer, - check_code_level=0, metric_key='u_f1', sampler=sampler, prefetch=True, use_tqdm=True, + check_code_level=0, metric_key='u_f1', sampler=sampler, num_workers=2, use_tqdm=True, device=device, callbacks=callbacks, update_every=update_every) trainer.train() \ No newline at end of file diff --git a/reproduction/matching/data/MatchingDataLoader.py b/reproduction/matching/data/MatchingDataLoader.py deleted file mode 100644 index bba26a8a..00000000 --- a/reproduction/matching/data/MatchingDataLoader.py +++ /dev/null @@ -1,435 +0,0 @@ -""" -这个文件的内容已合并到fastNLP.io.data_loader里，这个文件的内容不再更新 -""" - - -import os - -from typing import Union, Dict - -from fastNLP.core.const import Const -from fastNLP.core.vocabulary import Vocabulary -from fastNLP.io.base_loader import DataBundle, DataSetLoader -from fastNLP.io.dataset_loader import JsonLoader, CSVLoader -from fastNLP.io.file_utils import _get_base_url, cached_path, PRETRAINED_BERT_MODEL_DIR -from fastNLP.modules.encoder._bert import BertTokenizer - - -class MatchingLoader(DataSetLoader): - """ - 别名：:class:`fastNLP.io.MatchingLoader` :class:`fastNLP.io.dataset_loader.MatchingLoader` - - 读取Matching任务的数据集 - - :param dict paths: key是数据集名称（如train、dev、test），value是对应的文件名 - """ - - def __init__(self, paths: dict=None): - self.paths = paths - - def _load(self, path): - """ - :param str path: 待读取数据集的路径名 - :return: fastNLP.DataSet ds: 返回一个DataSet对象，里面必须包含3个field：其中两个分别为两个句子 - 的原始字符串文本，第三个为标签 - """ - raise NotImplementedError - - def process(self, paths: Union[str, Dict[str, str]], dataset_name: str=None, - to_lower=False, seq_len_type: str=None, bert_tokenizer: str=None, - cut_text: int = None, get_index=True, auto_pad_length: int=None, - auto_pad_token: str='', set_input: Union[list, str, bool]=True, - set_target: Union[list, str, bool] = True, concat: Union[str, list, bool]=None, ) -> DataBundle: - """ - :param paths: str或者Dict[str, str]。如果是str，则为数据集所在的文件夹或者是全路径文件名：如果是文件夹， - 则会从self.paths里面找对应的数据集名称与文件名。如果是Dict，则为数据集名称（如train、dev、test）和 - 对应的全路径文件名。 - :param str dataset_name: 如果在paths里传入的是一个数据集的全路径文件名，那么可以用dataset_name来定义 - 这个数据集的名字，如果不定义则默认为train。 - :param bool to_lower: 是否将文本自动转为小写。默认值为False。 - :param str seq_len_type: 提供的seq_len类型，支持 ``seq_len`` ：提供一个数字作为句子长度； ``mask`` : - 提供一个0/1的mask矩阵作为句子长度； ``bert`` ：提供segment_type_id（第一个句子为0，第二个句子为1）和 - attention mask矩阵（0/1的mask矩阵）。默认值为None，即不提供seq_len - :param str bert_tokenizer: bert tokenizer所使用的词表所在的文件夹路径 - :param int cut_text: 将长于cut_text的内容截掉。默认为None，即不截。 - :param bool get_index: 是否需要根据词表将文本转为index - :param int auto_pad_length: 是否需要将文本自动pad到一定长度（超过这个长度的文本将会被截掉），默认为不会自动pad - :param str auto_pad_token: 自动pad的内容 - :param set_input: 如果为True，则会自动将相关的field（名字里含有Const.INPUT的）设置为input，如果为False - 则不会将任何field设置为input。如果传入str或者List[str]，则会根据传入的内容将相对应的field设置为input， - 于此同时其他field不会被设置为input。默认值为True。 - :param set_target: set_target将控制哪些field可以被设置为target，用法与set_input一致。默认值为True。 - :param concat: 是否需要将两个句子拼接起来。如果为False则不会拼接。如果为True则会在两个句子之间插入一个。 - 如果传入一个长度为4的list，则分别表示插在第一句开始前、第一句结束后、第二句开始前、第二句结束后的标识符。如果 - 传入字符串 ``bert`` ，则会采用bert的拼接方式，等价于['[CLS]', '[SEP]', '', '[SEP]']. - :return: - """ - if isinstance(set_input, str): - set_input = [set_input] - if isinstance(set_target, str): - set_target = [set_target] - if isinstance(set_input, bool): - auto_set_input = set_input - else: - auto_set_input = False - if isinstance(set_target, bool): - auto_set_target = set_target - else: - auto_set_target = False - if isinstance(paths, str): - if os.path.isdir(paths): - path = {n: os.path.join(paths, self.paths[n]) for n in self.paths.keys()} - else: - path = {dataset_name if dataset_name is not None else 'train': paths} - else: - path = paths - - data_info = DataBundle() - for data_name in path.keys(): - data_info.datasets[data_name] = self._load(path[data_name]) - - for data_name, data_set in data_info.datasets.items(): - if auto_set_input: - data_set.set_input(Const.INPUTS(0), Const.INPUTS(1)) - if auto_set_target: - if Const.TARGET in data_set.get_field_names(): - data_set.set_target(Const.TARGET) - - if to_lower: - for data_name, data_set in data_info.datasets.items(): - data_set.apply(lambda x: [w.lower() for w in x[Const.INPUTS(0)]], new_field_name=Const.INPUTS(0), - is_input=auto_set_input) - data_set.apply(lambda x: [w.lower() for w in x[Const.INPUTS(1)]], new_field_name=Const.INPUTS(1), - is_input=auto_set_input) - - if bert_tokenizer is not None: - if bert_tokenizer.lower() in PRETRAINED_BERT_MODEL_DIR: - PRETRAIN_URL = _get_base_url('bert') - model_name = PRETRAINED_BERT_MODEL_DIR[bert_tokenizer] - model_url = PRETRAIN_URL + model_name - model_dir = cached_path(model_url) - # 检查是否存在 - elif os.path.isdir(bert_tokenizer): - model_dir = bert_tokenizer - else: - raise ValueError(f"Cannot recognize BERT tokenizer from {bert_tokenizer}.") - - words_vocab = Vocabulary(padding='[PAD]', unknown='[UNK]') - with open(os.path.join(model_dir, 'vocab.txt'), 'r') as f: - lines = f.readlines() - lines = [line.strip() for line in lines] - words_vocab.add_word_lst(lines) - words_vocab.build_vocab() - - tokenizer = BertTokenizer.from_pretrained(model_dir) - - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: tokenizer.tokenize(' '.join(x[fields])), new_field_name=fields, - is_input=auto_set_input) - - if isinstance(concat, bool): - concat = 'default' if concat else None - if concat is not None: - if isinstance(concat, str): - CONCAT_MAP = {'bert': ['[CLS]', '[SEP]', '', '[SEP]'], - 'default': ['', '', '', '']} - if concat.lower() in CONCAT_MAP: - concat = CONCAT_MAP[concat] - else: - concat = 4 * [concat] - assert len(concat) == 4, \ - f'Please choose a list with 4 symbols which at the beginning of first sentence ' \ - f'the end of first sentence, the begin of second sentence, and the end of second' \ - f'sentence. Your input is {concat}' - - for data_name, data_set in data_info.datasets.items(): - data_set.apply(lambda x: [concat[0]] + x[Const.INPUTS(0)] + [concat[1]] + [concat[2]] + - x[Const.INPUTS(1)] + [concat[3]], new_field_name=Const.INPUT) - data_set.apply(lambda x: [w for w in x[Const.INPUT] if len(w) > 0], new_field_name=Const.INPUT, - is_input=auto_set_input) - - if seq_len_type is not None: - if seq_len_type == 'seq_len': # - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: len(x[fields]), - new_field_name=fields.replace(Const.INPUT, Const.INPUT_LEN), - is_input=auto_set_input) - elif seq_len_type == 'mask': - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: [1] * len(x[fields]), - new_field_name=fields.replace(Const.INPUT, Const.INPUT_LEN), - is_input=auto_set_input) - elif seq_len_type == 'bert': - for data_name, data_set in data_info.datasets.items(): - if Const.INPUT not in data_set.get_field_names(): - raise KeyError(f'Field ``{Const.INPUT}`` not in {data_name} data set: ' - f'got {data_set.get_field_names()}') - data_set.apply(lambda x: [0] * (len(x[Const.INPUTS(0)]) + 2) + [1] * (len(x[Const.INPUTS(1)]) + 1), - new_field_name=Const.INPUT_LENS(0), is_input=auto_set_input) - data_set.apply(lambda x: [1] * len(x[Const.INPUT_LENS(0)]), - new_field_name=Const.INPUT_LENS(1), is_input=auto_set_input) - - if auto_pad_length is not None: - cut_text = min(auto_pad_length, cut_text if cut_text is not None else auto_pad_length) - - if cut_text is not None: - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if (Const.INPUT in fields) or ((Const.INPUT_LEN in fields) and (seq_len_type != 'seq_len')): - data_set.apply(lambda x: x[fields][: cut_text], new_field_name=fields, - is_input=auto_set_input) - - data_set_list = [d for n, d in data_info.datasets.items()] - assert len(data_set_list) > 0, f'There are NO data sets in data info!' - - if bert_tokenizer is None: - words_vocab = Vocabulary(padding=auto_pad_token) - words_vocab = words_vocab.from_dataset(*[d for n, d in data_info.datasets.items() if 'train' in n], - field_name=[n for n in data_set_list[0].get_field_names() - if (Const.INPUT in n)], - no_create_entry_dataset=[d for n, d in data_info.datasets.items() - if 'train' not in n]) - target_vocab = Vocabulary(padding=None, unknown=None) - target_vocab = target_vocab.from_dataset(*[d for n, d in data_info.datasets.items() if 'train' in n], - field_name=Const.TARGET) - data_info.vocabs = {Const.INPUT: words_vocab, Const.TARGET: target_vocab} - - if get_index: - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: [words_vocab.to_index(w) for w in x[fields]], new_field_name=fields, - is_input=auto_set_input) - - if Const.TARGET in data_set.get_field_names(): - data_set.apply(lambda x: target_vocab.to_index(x[Const.TARGET]), new_field_name=Const.TARGET, - is_input=auto_set_input, is_target=auto_set_target) - - if auto_pad_length is not None: - if seq_len_type == 'seq_len': - raise RuntimeError(f'the sequence will be padded with the length {auto_pad_length}, ' - f'so the seq_len_type cannot be `{seq_len_type}`!') - for data_name, data_set in data_info.datasets.items(): - for fields in data_set.get_field_names(): - if Const.INPUT in fields: - data_set.apply(lambda x: x[fields] + [words_vocab.to_index(words_vocab.padding)] * - (auto_pad_length - len(x[fields])), new_field_name=fields, - is_input=auto_set_input) - elif (Const.INPUT_LEN in fields) and (seq_len_type != 'seq_len'): - data_set.apply(lambda x: x[fields] + [0] * (auto_pad_length - len(x[fields])), - new_field_name=fields, is_input=auto_set_input) - - for data_name, data_set in data_info.datasets.items(): - if isinstance(set_input, list): - data_set.set_input(*[inputs for inputs in set_input if inputs in data_set.get_field_names()]) - if isinstance(set_target, list): - data_set.set_target(*[target for target in set_target if target in data_set.get_field_names()]) - - return data_info - - -class SNLILoader(MatchingLoader, JsonLoader): - """ - 别名：:class:`fastNLP.io.SNLILoader` :class:`fastNLP.io.dataset_loader.SNLILoader` - - 读取SNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - words2: list(str), 第二句文本, hypothesis - target: str, 真实标签 - - 数据来源: https://nlp.stanford.edu/projects/snli/snli_1.0.zip - """ - - def __init__(self, paths: dict=None): - fields = { - 'sentence1_binary_parse': Const.INPUTS(0), - 'sentence2_binary_parse': Const.INPUTS(1), - 'gold_label': Const.TARGET, - } - paths = paths if paths is not None else { - 'train': 'snli_1.0_train.jsonl', - 'dev': 'snli_1.0_dev.jsonl', - 'test': 'snli_1.0_test.jsonl'} - MatchingLoader.__init__(self, paths=paths) - JsonLoader.__init__(self, fields=fields) - - def _load(self, path): - ds = JsonLoader._load(self, path) - - parentheses_table = str.maketrans({'(': None, ')': None}) - - ds.apply(lambda ins: ins[Const.INPUTS(0)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(0)) - ds.apply(lambda ins: ins[Const.INPUTS(1)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(1)) - ds.drop(lambda x: x[Const.TARGET] == '-') - return ds - - -class RTELoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.RTELoader` :class:`fastNLP.io.dataset_loader.RTELoader` - - 读取RTE数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - words2: list(str), 第二句文本, hypothesis - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev': 'dev.tsv', - 'test': 'test.tsv' # test set has not label - } - MatchingLoader.__init__(self, paths=paths) - self.fields = { - 'sentence1': Const.INPUTS(0), - 'sentence2': Const.INPUTS(1), - 'label': Const.TARGET, - } - CSVLoader.__init__(self, sep='\t') - - def _load(self, path): - ds = CSVLoader._load(self, path) - - for k, v in self.fields.items(): - if v in ds.get_field_names(): - ds.rename_field(k, v) - for fields in ds.get_all_fields(): - if Const.INPUT in fields: - ds.apply(lambda x: x[fields].strip().split(), new_field_name=fields) - - return ds - - -class QNLILoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.QNLILoader` :class:`fastNLP.io.dataset_loader.QNLILoader` - - 读取QNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - words2: list(str), 第二句文本, hypothesis - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev': 'dev.tsv', - 'test': 'test.tsv' # test set has not label - } - MatchingLoader.__init__(self, paths=paths) - self.fields = { - 'question': Const.INPUTS(0), - 'sentence': Const.INPUTS(1), - 'label': Const.TARGET, - } - CSVLoader.__init__(self, sep='\t') - - def _load(self, path): - ds = CSVLoader._load(self, path) - - for k, v in self.fields.items(): - if v in ds.get_field_names(): - ds.rename_field(k, v) - for fields in ds.get_all_fields(): - if Const.INPUT in fields: - ds.apply(lambda x: x[fields].strip().split(), new_field_name=fields) - - return ds - - -class MNLILoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.MNLILoader` :class:`fastNLP.io.dataset_loader.MNLILoader` - - 读取MNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - words2: list(str), 第二句文本, hypothesis - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev_matched': 'dev_matched.tsv', - 'dev_mismatched': 'dev_mismatched.tsv', - 'test_matched': 'test_matched.tsv', - 'test_mismatched': 'test_mismatched.tsv', - # 'test_0.9_matched': 'multinli_0.9_test_matched_unlabeled.txt', - # 'test_0.9_mismatched': 'multinli_0.9_test_mismatched_unlabeled.txt', - - # test_0.9_mathed与mismatched是MNLI0.9版本的（数据来源：kaggle） - } - MatchingLoader.__init__(self, paths=paths) - CSVLoader.__init__(self, sep='\t') - self.fields = { - 'sentence1_binary_parse': Const.INPUTS(0), - 'sentence2_binary_parse': Const.INPUTS(1), - 'gold_label': Const.TARGET, - } - - def _load(self, path): - ds = CSVLoader._load(self, path) - - for k, v in self.fields.items(): - if k in ds.get_field_names(): - ds.rename_field(k, v) - - if Const.TARGET in ds.get_field_names(): - if ds[0][Const.TARGET] == 'hidden': - ds.delete_field(Const.TARGET) - - parentheses_table = str.maketrans({'(': None, ')': None}) - - ds.apply(lambda ins: ins[Const.INPUTS(0)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(0)) - ds.apply(lambda ins: ins[Const.INPUTS(1)].translate(parentheses_table).strip().split(), - new_field_name=Const.INPUTS(1)) - if Const.TARGET in ds.get_field_names(): - ds.drop(lambda x: x[Const.TARGET] == '-') - return ds - - -class QuoraLoader(MatchingLoader, CSVLoader): - """ - 别名：:class:`fastNLP.io.QuoraLoader` :class:`fastNLP.io.dataset_loader.QuoraLoader` - - 读取MNLI数据集，读取的DataSet包含fields:: - - words1: list(str)，第一句文本, premise - words2: list(str), 第二句文本, hypothesis - target: str, 真实标签 - - 数据来源: - """ - - def __init__(self, paths: dict=None): - paths = paths if paths is not None else { - 'train': 'train.tsv', - 'dev': 'dev.tsv', - 'test': 'test.tsv', - } - MatchingLoader.__init__(self, paths=paths) - CSVLoader.__init__(self, sep='\t', headers=(Const.TARGET, Const.INPUTS(0), Const.INPUTS(1), 'pairID')) - - def _load(self, path): - ds = CSVLoader._load(self, path) - return ds diff --git a/reproduction/matching/matching_bert.py b/reproduction/matching/matching_bert.py index 3ed75fd1..323d81a3 100644 --- a/reproduction/matching/matching_bert.py +++ b/reproduction/matching/matching_bert.py @@ -2,8 +2,12 @@ import random import numpy as np import torch -from fastNLP.core import Trainer, Tester, AccuracyMetric, Const, Adam -from fastNLP.io.data_loader import SNLILoader, RTELoader, MNLILoader, QNLILoader, QuoraLoader +from fastNLP.core import Trainer, Tester, AccuracyMetric, Const +from fastNLP.core.callback import WarmupCallback, EvaluateCallback +from fastNLP.core.optimizer import AdamW +from fastNLP.embeddings import BertEmbedding +from fastNLP.io.pipe.matching import SNLIBertPipe, RTEBertPipe, MNLIBertPipe,\ + QNLIBertPipe, QuoraBertPipe from reproduction.matching.model.bert import BertForNLI @@ -12,16 +16,22 @@ from reproduction.matching.model.bert import BertForNLI class BERTConfig: task = 'snli' + batch_size_per_gpu = 6 n_epochs = 6 lr = 2e-5 - seq_len_type = 'bert' + warm_up_rate = 0.1 seed = 42 + save_path = None # 模型存储的位置，None表示不存储模型。 + train_dataset_name = 'train' dev_dataset_name = 'dev' test_dataset_name = 'test' - save_path = None # 模型存储的位置，None表示不存储模型。 - bert_dir = 'path/to/bert/dir' # 预训练BERT参数文件的文件夹 + + to_lower = True # 忽略大小写 + tokenizer = 'spacy' # 使用spacy进行分词 + + bert_model_dir_or_name = 'bert-base-uncased' arg = BERTConfig() @@ -37,58 +47,52 @@ if n_gpu > 0: # load data set if arg.task == 'snli': - data_info = SNLILoader().process( - paths='path/to/snli/data', to_lower=True, seq_len_type=arg.seq_len_type, - bert_tokenizer=arg.bert_dir, cut_text=512, - get_index=True, concat='bert', - ) + data_bundle = SNLIBertPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'rte': - data_info = RTELoader().process( - paths='path/to/rte/data', to_lower=True, seq_len_type=arg.seq_len_type, - bert_tokenizer=arg.bert_dir, cut_text=512, - get_index=True, concat='bert', - ) + data_bundle = RTEBertPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'qnli': - data_info = QNLILoader().process( - paths='path/to/qnli/data', to_lower=True, seq_len_type=arg.seq_len_type, - bert_tokenizer=arg.bert_dir, cut_text=512, - get_index=True, concat='bert', - ) + data_bundle = QNLIBertPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'mnli': - data_info = MNLILoader().process( - paths='path/to/mnli/data', to_lower=True, seq_len_type=arg.seq_len_type, - bert_tokenizer=arg.bert_dir, cut_text=512, - get_index=True, concat='bert', - ) + data_bundle = MNLIBertPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'quora': - data_info = QuoraLoader().process( - paths='path/to/quora/data', to_lower=True, seq_len_type=arg.seq_len_type, - bert_tokenizer=arg.bert_dir, cut_text=512, - get_index=True, concat='bert', - ) + data_bundle = QuoraBertPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() else: raise RuntimeError(f'NOT support {arg.task} task yet!') +print(data_bundle) # print details in data_bundle + +# load embedding +embed = BertEmbedding(data_bundle.vocabs[Const.INPUT], model_dir_or_name=arg.bert_model_dir_or_name) + # define model -model = BertForNLI(class_num=len(data_info.vocabs[Const.TARGET]), bert_dir=arg.bert_dir) +model = BertForNLI(embed, class_num=len(data_bundle.vocabs[Const.TARGET])) + +# define optimizer and callback +optimizer = AdamW(lr=arg.lr, params=model.parameters()) +callbacks = [WarmupCallback(warmup=arg.warm_up_rate, schedule='linear'), ] + +if arg.task in ['snli']: + callbacks.append(EvaluateCallback(data=data_bundle.datasets[arg.test_dataset_name])) + # evaluate test set in every epoch if task is snli. # define trainer -trainer = Trainer(train_data=data_info.datasets[arg.train_dataset_name], model=model, - optimizer=Adam(lr=arg.lr, model_params=model.parameters()), +trainer = Trainer(train_data=data_bundle.datasets[arg.train_dataset_name], model=model, + optimizer=optimizer, batch_size=torch.cuda.device_count() * arg.batch_size_per_gpu, n_epochs=arg.n_epochs, print_every=-1, - dev_data=data_info.datasets[arg.dev_dataset_name], + dev_data=data_bundle.datasets[arg.dev_dataset_name], metrics=AccuracyMetric(), metric_key='acc', device=[i for i in range(torch.cuda.device_count())], check_code_level=-1, - save_path=arg.save_path) + save_path=arg.save_path, + callbacks=callbacks) # train model trainer.train(load_best_model=True) # define tester tester = Tester( - data=data_info.datasets[arg.test_dataset_name], + data=data_bundle.datasets[arg.test_dataset_name], model=model, metrics=AccuracyMetric(), batch_size=torch.cuda.device_count() * arg.batch_size_per_gpu, diff --git a/reproduction/matching/matching_cntn.py b/reproduction/matching/matching_cntn.py index 098f3bc4..9be716ba 100644 --- a/reproduction/matching/matching_cntn.py +++ b/reproduction/matching/matching_cntn.py @@ -1,9 +1,9 @@ import argparse import torch -from fastNLP.core import Trainer, Tester, Adam, AccuracyMetric, Const +from fastNLP.core import Trainer, Tester, Adam, AccuracyMetric, Const, CrossEntropyLoss from fastNLP.embeddings import StaticEmbedding -from fastNLP.io.data_loader import QNLILoader, RTELoader, SNLILoader, MNLILoader +from fastNLP.io.pipe.matching import SNLIPipe, RTEPipe, MNLIPipe, QNLIPipe from reproduction.matching.model.cntn import CNTNModel @@ -13,14 +13,12 @@ argument.add_argument('--embedding', choices=['glove', 'word2vec'], default='glo argument.add_argument('--batch-size-per-gpu', type=int, default=256) argument.add_argument('--n-epochs', type=int, default=200) argument.add_argument('--lr', type=float, default=1e-5) -argument.add_argument('--seq-len-type', choices=['mask', 'seq_len'], default='mask') argument.add_argument('--save-dir', type=str, default=None) argument.add_argument('--cntn-depth', type=int, default=1) argument.add_argument('--cntn-ns', type=int, default=200) argument.add_argument('--cntn-k-top', type=int, default=10) argument.add_argument('--cntn-r', type=int, default=5) argument.add_argument('--dataset', choices=['qnli', 'rte', 'snli', 'mnli'], default='qnli') -argument.add_argument('--max-len', type=int, default=50) arg = argument.parse_args() # dataset dict @@ -45,30 +43,25 @@ else: num_labels = 3 # load data set -if arg.dataset == 'qnli': - data_info = QNLILoader().process( - paths='path/to/qnli/data', to_lower=True, seq_len_type=arg.seq_len_type, bert_tokenizer=None, - get_index=True, concat=False, auto_pad_length=arg.max_len) +if arg.dataset == 'snli': + data_bundle = SNLIPipe(lower=True, tokenizer='raw').process_from_file() elif arg.dataset == 'rte': - data_info = RTELoader().process( - paths='path/to/rte/data', to_lower=True, seq_len_type=arg.seq_len_type, bert_tokenizer=None, - get_index=True, concat=False, auto_pad_length=arg.max_len) -elif arg.dataset == 'snli': - data_info = SNLILoader().process( - paths='path/to/snli/data', to_lower=True, seq_len_type=arg.seq_len_type, bert_tokenizer=None, - get_index=True, concat=False, auto_pad_length=arg.max_len) + data_bundle = RTEPipe(lower=True, tokenizer='raw').process_from_file() +elif arg.dataset == 'qnli': + data_bundle = QNLIPipe(lower=True, tokenizer='raw').process_from_file() elif arg.dataset == 'mnli': - data_info = MNLILoader().process( - paths='path/to/mnli/data', to_lower=True, seq_len_type=arg.seq_len_type, bert_tokenizer=None, - get_index=True, concat=False, auto_pad_length=arg.max_len) + data_bundle = MNLIPipe(lower=True, tokenizer='raw').process_from_file() else: - raise ValueError(f'now we only support [qnli,rte,snli,mnli] dataset for cntn model!') + raise RuntimeError(f'NOT support {arg.task} task yet!') + +print(data_bundle) # print details in data_bundle # load embedding if arg.embedding == 'word2vec': - embedding = StaticEmbedding(data_info.vocabs[Const.INPUT], model_dir_or_name='en-word2vec-300', requires_grad=True) + embedding = StaticEmbedding(data_bundle.vocabs[Const.INPUTS(0)], model_dir_or_name='en-word2vec-300', + requires_grad=True) elif arg.embedding == 'glove': - embedding = StaticEmbedding(data_info.vocabs[Const.INPUT], model_dir_or_name='en-glove-840b-300', + embedding = StaticEmbedding(data_bundle.vocabs[Const.INPUTS(0)], model_dir_or_name='en-glove-840b-300d', requires_grad=True) else: raise ValueError(f'now we only support word2vec or glove embedding for cntn model!') @@ -79,11 +72,12 @@ model = CNTNModel(embedding, ns=arg.cntn_ns, k_top=arg.cntn_k_top, num_labels=nu print(model) # define trainer -trainer = Trainer(train_data=data_info.datasets['train'], model=model, +trainer = Trainer(train_data=data_bundle.datasets['train'], model=model, optimizer=Adam(lr=arg.lr, model_params=model.parameters()), + loss=CrossEntropyLoss(), batch_size=torch.cuda.device_count() * arg.batch_size_per_gpu, n_epochs=arg.n_epochs, print_every=-1, - dev_data=data_info.datasets[dev_dict[arg.dataset]], + dev_data=data_bundle.datasets[dev_dict[arg.dataset]], metrics=AccuracyMetric(), metric_key='acc', device=[i for i in range(torch.cuda.device_count())], check_code_level=-1) @@ -93,7 +87,7 @@ trainer.train(load_best_model=True) # define tester tester = Tester( - data=data_info.datasets[test_dict[arg.dataset]], + data=data_bundle.datasets[test_dict[arg.dataset]], model=model, metrics=AccuracyMetric(), batch_size=torch.cuda.device_count() * arg.batch_size_per_gpu, diff --git a/reproduction/matching/matching_esim.py b/reproduction/matching/matching_esim.py index 2ff6916a..9d50c0fb 100644 --- a/reproduction/matching/matching_esim.py +++ b/reproduction/matching/matching_esim.py @@ -6,10 +6,11 @@ from torch.optim import Adamax from torch.optim.lr_scheduler import StepLR from fastNLP.core import Trainer, Tester, AccuracyMetric, Const -from fastNLP.core.callback import GradientClipCallback, LRScheduler -from fastNLP.embeddings.static_embedding import StaticEmbedding -from fastNLP.embeddings.elmo_embedding import ElmoEmbedding -from fastNLP.io.data_loader import SNLILoader, RTELoader, MNLILoader, QNLILoader, QuoraLoader +from fastNLP.core.callback import GradientClipCallback, LRScheduler, EvaluateCallback +from fastNLP.core.losses import CrossEntropyLoss +from fastNLP.embeddings import StaticEmbedding +from fastNLP.embeddings import ElmoEmbedding +from fastNLP.io.pipe.matching import SNLIPipe, RTEPipe, MNLIPipe, QNLIPipe, QuoraPipe from fastNLP.models.snli import ESIM @@ -17,18 +18,21 @@ from fastNLP.models.snli import ESIM class ESIMConfig: task = 'snli' + embedding = 'glove' + batch_size_per_gpu = 196 n_epochs = 30 lr = 2e-3 - seq_len_type = 'seq_len' - # seq_len表示在process的时候用len(words)来表示长度信息； - # mask表示用0/1掩码矩阵来表示长度信息； seed = 42 + save_path = None # 模型存储的位置，None表示不存储模型。 + train_dataset_name = 'train' dev_dataset_name = 'dev' test_dataset_name = 'test' - save_path = None # 模型存储的位置，None表示不存储模型。 + + to_lower = True # 忽略大小写 + tokenizer = 'spacy' # 使用spacy进行分词 arg = ESIMConfig() @@ -44,43 +48,32 @@ if n_gpu > 0: # load data set if arg.task == 'snli': - data_info = SNLILoader().process( - paths='path/to/snli/data', to_lower=False, seq_len_type=arg.seq_len_type, - get_index=True, concat=False, - ) + data_bundle = SNLIPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'rte': - data_info = RTELoader().process( - paths='path/to/rte/data', to_lower=False, seq_len_type=arg.seq_len_type, - get_index=True, concat=False, - ) + data_bundle = RTEPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'qnli': - data_info = QNLILoader().process( - paths='path/to/qnli/data', to_lower=False, seq_len_type=arg.seq_len_type, - get_index=True, concat=False, - ) + data_bundle = QNLIPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'mnli': - data_info = MNLILoader().process( - paths='path/to/mnli/data', to_lower=False, seq_len_type=arg.seq_len_type, - get_index=True, concat=False, - ) + data_bundle = MNLIPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() elif arg.task == 'quora': - data_info = QuoraLoader().process( - paths='path/to/quora/data', to_lower=False, seq_len_type=arg.seq_len_type, - get_index=True, concat=False, - ) + data_bundle = QuoraPipe(lower=arg.to_lower, tokenizer=arg.tokenizer).process_from_file() else: raise RuntimeError(f'NOT support {arg.task} task yet!') +print(data_bundle) # print details in data_bundle + # load embedding if arg.embedding == 'elmo': - embedding = ElmoEmbedding(data_info.vocabs[Const.INPUT], requires_grad=True) + embedding = ElmoEmbedding(data_bundle.vocabs[Const.INPUTS(0)], model_dir_or_name='en-medium', + requires_grad=True) elif arg.embedding == 'glove': - embedding = StaticEmbedding(data_info.vocabs[Const.INPUT], requires_grad=True, normalize=False) + embedding = StaticEmbedding(data_bundle.vocabs[Const.INPUTS(0)], model_dir_or_name='en-glove-840b-300d', + requires_grad=True, normalize=False) else: raise RuntimeError(f'NOT support {arg.embedding} embedding yet!') # define model -model = ESIM(embedding, num_labels=len(data_info.vocabs[Const.TARGET])) +model = ESIM(embedding, num_labels=len(data_bundle.vocabs[Const.TARGET])) # define optimizer and callback optimizer = Adamax(lr=arg.lr, params=model.parameters()) @@ -91,23 +84,29 @@ callbacks = [ LRScheduler(scheduler), ] +if arg.task in ['snli']: + callbacks.append(EvaluateCallback(data=data_bundle.datasets[arg.test_dataset_name])) + # evaluate test set in every epoch if task is snli. + # define trainer -trainer = Trainer(train_data=data_info.datasets[arg.train_dataset_name], model=model, +trainer = Trainer(train_data=data_bundle.datasets[arg.train_dataset_name], model=model, optimizer=optimizer, + loss=CrossEntropyLoss(), batch_size=torch.cuda.device_count() * arg.batch_size_per_gpu, n_epochs=arg.n_epochs, print_every=-1, - dev_data=data_info.datasets[arg.dev_dataset_name], + dev_data=data_bundle.datasets[arg.dev_dataset_name], metrics=AccuracyMetric(), metric_key='acc', device=[i for i in range(torch.cuda.device_count())], check_code_level=-1, - save_path=arg.save_path) + save_path=arg.save_path, + callbacks=callbacks) # train model trainer.train(load_best_model=True) # define tester tester = Tester( - data=data_info.datasets[arg.test_dataset_name], + data=data_bundle.datasets[arg.test_dataset_name], model=model, metrics=AccuracyMetric(), batch_size=torch.cuda.device_count() * arg.batch_size_per_gpu, diff --git a/reproduction/matching/matching_mwan.py b/reproduction/matching/matching_mwan.py index 31af54c5..026ea7b4 100644 --- a/reproduction/matching/matching_mwan.py +++ b/reproduction/matching/matching_mwan.py @@ -6,12 +6,11 @@ from torch.optim import Adadelta from torch.optim.lr_scheduler import StepLR from fastNLP import CrossEntropyLoss -from fastNLP import cache_results from fastNLP.core import Trainer, Tester, AccuracyMetric, Const -from fastNLP.core.callback import LRScheduler, FitlogCallback +from fastNLP.core.callback import LRScheduler, EvaluateCallback from fastNLP.embeddings import StaticEmbedding -from fastNLP.io.data_loader import MNLILoader, QNLILoader, SNLILoader, RTELoader +from fastNLP.io.pipe.matching import SNLIPipe, RTEPipe, MNLIPipe, QNLIPipe, QuoraPipe from reproduction.matching.model.mwan import MwanModel import fitlog @@ -46,47 +45,25 @@ for k in arg.__dict__: # load data set if arg.task == 'snli': - @cache_results(f'snli_mwan.pkl') - def read_snli(): - data_info = SNLILoader().process( - paths='path/to/snli/data', to_lower=True, seq_len_type=None, bert_tokenizer=None, - get_index=True, concat=False, extra_split=['/','%','-'], - ) - return data_info - data_info = read_snli() + data_bundle = SNLIPipe(lower=True, tokenizer='spacy').process_from_file() elif arg.task == 'rte': - @cache_results(f'rte_mwan.pkl') - def read_rte(): - data_info = RTELoader().process( - paths='path/to/rte/data', to_lower=True, seq_len_type=None, bert_tokenizer=None, - get_index=True, concat=False, extra_split=['/','%','-'], - ) - return data_info - data_info = read_rte() + data_bundle = RTEPipe(lower=True, tokenizer='spacy').process_from_file() elif arg.task == 'qnli': - data_info = QNLILoader().process( - paths='path/to/qnli/data', to_lower=True, seq_len_type=None, bert_tokenizer=None, - get_index=True, concat=False , cut_text=512, extra_split=['/','%','-'], - ) + data_bundle = QNLIPipe(lower=True, tokenizer='spacy').process_from_file() elif arg.task == 'mnli': - @cache_results(f'mnli_v0.9_mwan.pkl') - def read_mnli(): - data_info = MNLILoader().process( - paths='path/to/mnli/data', to_lower=True, seq_len_type=None, bert_tokenizer=None, - get_index=True, concat=False, extra_split=['/','%','-'], - ) - return data_info - data_info = read_mnli() + data_bundle = MNLIPipe(lower=True, tokenizer='spacy').process_from_file() +elif arg.task == 'quora': + data_bundle = QuoraPipe(lower=True, tokenizer='spacy').process_from_file() else: raise RuntimeError(f'NOT support {arg.task} task yet!') -print(data_info) -print(len(data_info.vocabs['words'])) +print(data_bundle) +print(len(data_bundle.vocabs[Const.INPUTS(0)])) model = MwanModel( - num_class = len(data_info.vocabs[Const.TARGET]), - EmbLayer = StaticEmbedding(data_info.vocabs[Const.INPUT], requires_grad=False, normalize=False), + num_class = len(data_bundle.vocabs[Const.TARGET]), + EmbLayer = StaticEmbedding(data_bundle.vocabs[Const.INPUTS(0)], requires_grad=False, normalize=False), ElmoLayer = None, args_of_imm = { "input_size" : 300 , @@ -105,21 +82,20 @@ callbacks = [ ] if arg.task in ['snli']: - callbacks.append(FitlogCallback(data_info.datasets[arg.testset_name], verbose=1)) + callbacks.append(EvaluateCallback(data=data_bundle.datasets[arg.testset_name])) elif arg.task == 'mnli': - callbacks.append(FitlogCallback({'dev_matched': data_info.datasets['dev_matched'], - 'dev_mismatched': data_info.datasets['dev_mismatched']}, - verbose=1)) + callbacks.append(EvaluateCallback(data={'dev_matched': data_bundle.datasets['dev_matched'], + 'dev_mismatched': data_bundle.datasets['dev_mismatched']},)) trainer = Trainer( - train_data = data_info.datasets['train'], + train_data = data_bundle.datasets['train'], model = model, optimizer = optimizer, num_workers = 0, batch_size = arg.batch_size, n_epochs = arg.n_epochs, print_every = -1, - dev_data = data_info.datasets[arg.devset_name], + dev_data = data_bundle.datasets[arg.devset_name], metrics = AccuracyMetric(pred = "pred" , target = "target"), metric_key = 'acc', device = [i for i in range(torch.cuda.device_count())], @@ -130,7 +106,7 @@ trainer = Trainer( trainer.train(load_best_model=True) tester = Tester( - data=data_info.datasets[arg.testset_name], + data=data_bundle.datasets[arg.testset_name], model=model, metrics=AccuracyMetric(), batch_size=arg.batch_size, diff --git a/reproduction/matching/model/bert.py b/reproduction/matching/model/bert.py index a21f8c36..73a0c533 100644 --- a/reproduction/matching/model/bert.py +++ b/reproduction/matching/model/bert.py @@ -3,39 +3,28 @@ import torch import torch.nn as nn from fastNLP.core.const import Const -from fastNLP.models import BaseModel -from fastNLP.embeddings.bert import BertModel +from fastNLP.models.base_model import BaseModel +from fastNLP.embeddings import BertEmbedding class BertForNLI(BaseModel): - # TODO: still in progress - def __init__(self, class_num=3, bert_dir=None): + def __init__(self, bert_embed: BertEmbedding, class_num=3): super(BertForNLI, self).__init__() - if bert_dir is not None: - self.bert = BertModel.from_pretrained(bert_dir) - else: - self.bert = BertModel() - hidden_size = self.bert.pooler.dense._parameters['bias'].size(-1) - self.classifier = nn.Linear(hidden_size, class_num) - - def forward(self, words, seq_len1, seq_len2, target=None): + self.embed = bert_embed + self.classifier = nn.Linear(self.embed.embedding_dim, class_num) + + def forward(self, words): """ :param torch.Tensor words: [batch_size, seq_len] input_ids - :param torch.Tensor seq_len1: [batch_size, seq_len] token_type_ids - :param torch.Tensor seq_len2: [batch_size, seq_len] attention_mask - :param torch.Tensor target: [batch] :return: """ - _, pooled_output = self.bert(words, seq_len1, seq_len2) - logits = self.classifier(pooled_output) + hidden = self.embed(words) + logits = self.classifier(hidden) - if target is not None: - loss_func = torch.nn.CrossEntropyLoss() - loss = loss_func(logits, target) - return {Const.OUTPUT: logits, Const.LOSS: loss} return {Const.OUTPUT: logits} - def predict(self, words, seq_len1, seq_len2, target=None): - return self.forward(words, seq_len1, seq_len2) + def predict(self, words): + logits = self.forward(words)[Const.OUTPUT] + return {Const.OUTPUT: logits.argmax(dim=-1)} diff --git a/reproduction/matching/model/cntn.py b/reproduction/matching/model/cntn.py index a0a104a3..cfa5e5a8 100644 --- a/reproduction/matching/model/cntn.py +++ b/reproduction/matching/model/cntn.py @@ -3,10 +3,8 @@ import torch.nn as nn import torch.nn.functional as F import numpy as np -from torch.nn import CrossEntropyLoss - -from fastNLP.models import BaseModel -from fastNLP.embeddings.embedding import TokenEmbedding +from fastNLP.models.base_model import BaseModel +from fastNLP.embeddings import TokenEmbedding from fastNLP.core.const import Const @@ -83,13 +81,12 @@ class CNTNModel(BaseModel): self.weight_V = nn.Linear(2 * ns, r) self.weight_u = nn.Sequential(nn.Dropout(p=dropout_rate), nn.Linear(r, num_labels)) - def forward(self, words1, words2, seq_len1, seq_len2, target=None): + def forward(self, words1, words2, seq_len1, seq_len2): """ :param words1: [batch, seq_len, emb_size] Question. :param words2: [batch, seq_len, emb_size] Answer. :param seq_len1: [batch] :param seq_len2: [batch] - :param target: [batch] Glod labels. :return: """ in_q = self.embedding(words1) @@ -109,12 +106,7 @@ class CNTNModel(BaseModel): in_a = self.fc_q(in_a.view(in_a.size(0), -1)) score = torch.tanh(self.weight_u(self.weight_M(in_q, in_a) + self.weight_V(torch.cat((in_q, in_a), -1)))) - if target is not None: - loss_fct = CrossEntropyLoss() - loss = loss_fct(score, target) - return {Const.LOSS: loss, Const.OUTPUT: score} - else: - return {Const.OUTPUT: score} + return {Const.OUTPUT: score} - def predict(self, **kwargs): - return self.forward(**kwargs) + def predict(self, words1, words2, seq_len1, seq_len2): + return self.forward(words1, words2, seq_len1, seq_len2) diff --git a/reproduction/matching/model/esim.py b/reproduction/matching/model/esim.py index 87e5ba65..d704e2f8 100644 --- a/reproduction/matching/model/esim.py +++ b/reproduction/matching/model/esim.py @@ -2,10 +2,8 @@ import torch import torch.nn as nn import torch.nn.functional as F -from torch.nn import CrossEntropyLoss - -from fastNLP.models import BaseModel -from fastNLP.embeddings.embedding import TokenEmbedding +from fastNLP.models.base_model import BaseModel +from fastNLP.embeddings import TokenEmbedding from fastNLP.core.const import Const from fastNLP.core.utils import seq_len_to_mask @@ -42,13 +40,12 @@ class ESIMModel(BaseModel): nn.init.xavier_uniform_(self.classifier[1].weight.data) nn.init.xavier_uniform_(self.classifier[4].weight.data) - def forward(self, words1, words2, seq_len1, seq_len2, target=None): + def forward(self, words1, words2, seq_len1, seq_len2): """ :param words1: [batch, seq_len] :param words2: [batch, seq_len] :param seq_len1: [batch] :param seq_len2: [batch] - :param target: :return: """ mask1 = seq_len_to_mask(seq_len1, words1.size(1)) @@ -82,16 +79,10 @@ class ESIMModel(BaseModel): logits = torch.tanh(self.classifier(out)) # logits = self.classifier(out) - if target is not None: - loss_fct = CrossEntropyLoss() - loss = loss_fct(logits, target) - - return {Const.LOSS: loss, Const.OUTPUT: logits} - else: - return {Const.OUTPUT: logits} + return {Const.OUTPUT: logits} - def predict(self, **kwargs): - pred = self.forward(**kwargs)[Const.OUTPUT].argmax(-1) + def predict(self, words1, words2, seq_len1, seq_len2): + pred = self.forward(words1, words2, seq_len1, seq_len2)[Const.OUTPUT].argmax(-1) return {Const.OUTPUT: pred} # input [batch_size, len , hidden] diff --git a/reproduction/matching/test/test_snlidataloader.py b/reproduction/matching/test/test_snlidataloader.py deleted file mode 100644 index 60b3ad59..00000000 --- a/reproduction/matching/test/test_snlidataloader.py +++ /dev/null @@ -1,10 +0,0 @@ -import unittest -from ..data import MatchingDataLoader -from fastNLP.core.vocabulary import Vocabulary - - -class TestCWSDataLoader(unittest.TestCase): - def test_case1(self): - snli_loader = MatchingDataLoader() - # TODO: still in progress - diff --git a/reproduction/seqence_labelling/chinese_ner/data/ChineseNER.py b/reproduction/seqence_labelling/chinese_ner/data/ChineseNER.py deleted file mode 100644 index cec5ab76..00000000 --- a/reproduction/seqence_labelling/chinese_ner/data/ChineseNER.py +++ /dev/null @@ -1,115 +0,0 @@ - - -from fastNLP.io.base_loader import DataSetLoader, DataBundle -from fastNLP.io import ConllLoader -from reproduction.seqence_labelling.ner.data.utils import iob2bioes, iob2 -from fastNLP import Const -from reproduction.utils import check_dataloader_paths -from fastNLP import Vocabulary - -class ChineseNERLoader(DataSetLoader): - """ - 读取中文命名实体数据集，包括PeopleDaily, MSRA-NER, Weibo。数据在这里可以找到https://github.com/OYE93/Chinese-NLP-Corpus/tree/master/NER - 请确保输入数据的格式如下, 共两列，第一列为字，第二列为标签，不同句子以空行隔开 - 我 O - 们 O - 变 O - 而 O - 以 O - 书 O - 会 O - ... - - """ - def __init__(self, encoding_type:str='bioes'): - """ - - :param str encoding_type: 支持bio和bioes格式 - """ - super().__init__() - self._loader = ConllLoader(headers=['raw_chars', 'target'], indexes=[0, 1]) - - assert encoding_type in ('bio', 'bioes') - - self._tag_converters = [iob2] - if encoding_type == 'bioes': - self._tag_converters.append(iob2bioes) - - def load(self, path:str): - dataset = self._loader.load(path) - def convert_tag_schema(tags): - for converter in self._tag_converters: - tags = converter(tags) - return tags - if self._tag_converters: - dataset.apply_field(convert_tag_schema, field_name=Const.TARGET, new_field_name=Const.TARGET) - return dataset - - def process(self, paths, bigrams=False, trigrams=False): - """ - - :param paths: - :param bool, bigrams: 是否包含生成bigram feature, [a, b, c, d] -> [ab, bc, cd, d] - :param bool, trigrams: 是否包含trigram feature，[a, b, c, d] -> [abc, bcd, cd, d] - :return: DataBundle - 包含以下的fields - raw_chars: List[str] - chars: List[int] - seq_len: int, 字的长度 - bigrams: List[int], optional - trigrams: List[int], optional - target: List[int] - """ - paths = check_dataloader_paths(paths) - data = DataBundle() - input_fields = [Const.CHAR_INPUT, Const.INPUT_LEN, Const.TARGET] - target_fields = [Const.TARGET, Const.INPUT_LEN] - - for name, path in paths.items(): - dataset = self.load(path) - if bigrams: - dataset.apply_field(lambda raw_chars: [c1+c2 for c1, c2 in zip(raw_chars, raw_chars[1:]+[''])], - field_name='raw_chars', new_field_name='bigrams') - - if trigrams: - dataset.apply_field(lambda raw_chars: [c1+c2+c3 for c1, c2, c3 in zip(raw_chars, - raw_chars[1:]+[''], - raw_chars[2:]+['']*2)], - field_name='raw_chars', new_field_name='trigrams') - data.datasets[name] = dataset - - char_vocab = Vocabulary().from_dataset(data.datasets['train'], field_name='raw_chars', - no_create_entry_dataset=[dataset for name, dataset in data.datasets.items() if name!='train']) - char_vocab.index_dataset(*data.datasets.values(), field_name='raw_chars', new_field_name=Const.CHAR_INPUT) - data.vocabs[Const.CHAR_INPUT] = char_vocab - - target_vocab = Vocabulary(unknown=None, padding=None).from_dataset(data.datasets['train'], field_name=Const.TARGET) - target_vocab.index_dataset(*data.datasets.values(), field_name=Const.TARGET) - data.vocabs[Const.TARGET] = target_vocab - - if bigrams: - bigram_vocab = Vocabulary().from_dataset(data.datasets['train'], field_name='bigrams', - no_create_entry_dataset=[dataset for name, dataset in - data.datasets.items() if name != 'train']) - bigram_vocab.index_dataset(*data.datasets.values(), field_name='bigrams', new_field_name='bigrams') - data.vocabs['bigrams'] = bigram_vocab - input_fields.append('bigrams') - - if trigrams: - trigram_vocab = Vocabulary().from_dataset(data.datasets['train'], field_name='trigrams', - no_create_entry_dataset=[dataset for name, dataset in - data.datasets.items() if name != 'train']) - trigram_vocab.index_dataset(*data.datasets.values(), field_name='trigrams', new_field_name='trigrams') - data.vocabs['trigrams'] = trigram_vocab - input_fields.append('trigrams') - - for name, dataset in data.datasets.items(): - dataset.add_seq_len(Const.CHAR_INPUT) - dataset.set_input(*input_fields) - dataset.set_target(*target_fields) - - return data - - - - diff --git a/reproduction/seqence_labelling/chinese_ner/data/__init__.py b/reproduction/seqence_labelling/chinese_ner/data/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/reproduction/seqence_labelling/chinese_ner/readme.md b/reproduction/seqence_labelling/chinese_ner/readme.md new file mode 100644 index 00000000..3a9d37d8 --- /dev/null +++ b/reproduction/seqence_labelling/chinese_ner/readme.md @@ -0,0 +1,30 @@ +使用以下中文NERPipe自动下载的统计数据 + +| MsraNERPipe | # of sents | # of tokens | +| ----------- | ---------- | ----------- | +| train | 41747 | 1954374 | +| dev | 4617 | 215505 | +| test | 4365 | 172601 | +| total | 50729 | 2342480 | +这里报道的统计数据，与[https://arxiv.org/pdf/1805.02023.pdf]()报道的一致 + + + +| WeiboNERPipe | # of sents | # of tokens | +| ------------ | ---------- | ----------- | +| train | 1350 | 73778 | +| dev | 270 | 14509 | +| test | 270 | 14842 | +| total | 1890 | 1890 | +这里报道的统计数据与[https://www.cs.cmu.edu/~ark/EMNLP-2015/proceedings/EMNLP/pdf/EMNLP064.pdf]()一致 + + + + +| PeopleDailyPipe | # of sents | # of tokens | +| --------------- | ---------- | ----------- | +| train | 50658 | 2169879 | +| dev | 4631 | 172601 | +| test | 68 | 2270 | +| total | 55357 | 2344750 | +这里使用的数据与[https://arxiv.org/pdf/1906.08101.pdf]()的数据是一致的 diff --git a/reproduction/seqence_labelling/chinese_ner/train_bert.py b/reproduction/seqence_labelling/chinese_ner/train_bert.py index a34b7d01..b12c8f75 100644 --- a/reproduction/seqence_labelling/chinese_ner/train_bert.py +++ b/reproduction/seqence_labelling/chinese_ner/train_bert.py @@ -12,22 +12,23 @@ sys.path.append('../../../') from torch import nn from fastNLP.embeddings import BertEmbedding, Embedding -from reproduction.seqence_labelling.chinese_ner.data.ChineseNER import ChineseNERLoader from fastNLP import Trainer, Const from fastNLP import BucketSampler, SpanFPreRecMetric, GradientClipCallback from fastNLP.modules import MLP from fastNLP.core.callback import WarmupCallback from fastNLP import CrossEntropyLoss from fastNLP.core.optimizer import AdamW -import os +from fastNLP.io import MsraNERPipe, MsraNERLoader, WeiboNERPipe from fastNLP import cache_results encoding_type = 'bio' -@cache_results('caches/msra.pkl') +@cache_results('caches/weibo.pkl', _refresh=False) def get_data(): - data = ChineseNERLoader(encoding_type=encoding_type).process("MSRA/") + # data_dir = MsraNERLoader().download(dev_ratio=0) + # data = MsraNERPipe(encoding_type=encoding_type, target_pad_val=-100).process_from_file(data_dir) + data = WeiboNERPipe(encoding_type=encoding_type).process_from_file() return data data = get_data() print(data) @@ -35,10 +36,10 @@ print(data) class BertCNNER(nn.Module): def __init__(self, embed, tag_size): super().__init__() - - self.embedding = Embedding(embed, dropout=0.1) + self.embedding = embed self.tag_size = tag_size self.mlp = MLP(size_layer=[self.embedding.embedding_dim, tag_size]) + def forward(self, chars): # batch_size, max_len = words.size() chars = self.embedding(chars) @@ -46,11 +47,15 @@ class BertCNNER(nn.Module): return {Const.OUTPUT: outputs} -embed = BertEmbedding(data.vocabs[Const.CHAR_INPUT], model_dir_or_name='en-base', - pool_method='max', requires_grad=True, layers='11') + def predict(self, chars): + # batch_size, max_len = words.size() + chars = self.embedding(chars) + outputs = self.mlp(chars) -for name, dataset in data.datasets.items(): - dataset.set_pad_val(Const.TARGET, -100) + return {Const.OUTPUT: outputs} + +embed = BertEmbedding(data.get_vocab(Const.CHAR_INPUT), model_dir_or_name='cn-wwm-ext', + pool_method='first', requires_grad=True, layers='11', include_cls_sep=False, dropout=0.5) callbacks = [ GradientClipCallback(clip_type='norm', clip_value=1), @@ -58,7 +63,7 @@ callbacks = [ ] model = BertCNNER(embed, len(data.vocabs[Const.TARGET])) -optimizer = AdamW(model.parameters(), lr=1e-4) +optimizer = AdamW(model.parameters(), lr=3e-5) for name, dataset in data.datasets.items(): original_len = len(dataset) @@ -66,13 +71,11 @@ for name, dataset in data.datasets.items(): clipped_len = len(dataset) print("Delete {} instances in {}.".format(original_len-clipped_len, name)) -os.environ['CUDA_VISIBLE_DEVICES'] = '0,1' - trainer = Trainer(train_data=data.datasets['train'], model=model, optimizer=optimizer, sampler=BucketSampler(), - device=[0, 1], dev_data=data.datasets['test'], batch_size=20, + device=0, dev_data=data.datasets['test'], batch_size=6, metrics=SpanFPreRecMetric(tag_vocab=data.vocabs[Const.TARGET], encoding_type=encoding_type), loss=CrossEntropyLoss(reduction='sum'), callbacks=callbacks, num_workers=2, n_epochs=5, - check_code_level=-1, update_every=3) + check_code_level=0, update_every=3) trainer.train() diff --git a/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py b/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py index 53a85186..58b32265 100644 --- a/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py +++ b/reproduction/seqence_labelling/chinese_ner/train_cn_ner.py @@ -1,7 +1,6 @@ +import sys +sys.path.append('../../..') - - -from reproduction.seqence_labelling.chinese_ner.data.ChineseNER import ChineseNERLoader from fastNLP.embeddings import StaticEmbedding from torch import nn @@ -14,7 +13,51 @@ import torch.nn.functional as F from fastNLP import seq_len_to_mask from fastNLP.core.const import Const as C from fastNLP import SpanFPreRecMetric, Trainer -from fastNLP import cache_results +from fastNLP import cache_results, Vocabulary +from fastNLP.io.pipe.utils import _add_chars_field, _indexize + +from fastNLP.io.pipe import Pipe +from fastNLP.core.utils import iob2bioes, iob2 +from fastNLP.io import MsraNERLoader, WeiboNERLoader + +class ChineseNERPipe(Pipe): + def __init__(self, encoding_type: str = 'bio', target_pad_val=0, bigram=False): + if encoding_type == 'bio': + self.convert_tag = iob2 + else: + self.convert_tag = lambda words: iob2bioes(iob2(words)) + self.target_pad_val = int(target_pad_val) + self.bigram = bigram + + def process(self, data_bundle): + data_bundle.copy_field(C.RAW_CHAR, C.CHAR_INPUT) + input_fields = [C.TARGET, C.CHAR_INPUT, C.INPUT_LEN] + target_fields = [C.TARGET, C.INPUT_LEN] + if self.bigram: + for dataset in data_bundle.datasets.values(): + dataset.apply_field(lambda chars:[c1+c2 for c1, c2 in zip(chars, chars[1:]+[''])], + field_name=C.CHAR_INPUT, new_field_name='bigrams') + bigram_vocab = Vocabulary() + bigram_vocab.from_dataset(data_bundle.get_dataset('train'),field_name='bigrams', + no_create_entry_dataset=[ds for name, ds in data_bundle.datasets.items() if name!='train']) + bigram_vocab.index_dataset(*data_bundle.datasets.values(), field_name='bigrams') + data_bundle.set_vocab(bigram_vocab, field_name='bigrams') + input_fields.append('bigrams') + + _add_chars_field(data_bundle, lower=False) + + # index + _indexize(data_bundle, input_field_names=C.CHAR_INPUT, target_field_names=C.TARGET) + + for name, dataset in data_bundle.datasets.items(): + dataset.set_pad_val(C.TARGET, self.target_pad_val) + dataset.add_seq_len(C.CHAR_INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + class CNBiLSTMCRFNER(nn.Module): def __init__(self, char_embed, num_classes, bigram_embed=None, trigram_embed=None, num_layers=1, hidden_size=100, @@ -73,22 +116,21 @@ class CNBiLSTMCRFNER(nn.Module): return self._forward(chars, bigrams, trigrams, seq_len) # data_bundle = pickle.load(open('caches/msra.pkl', 'rb')) -@cache_results('caches/msra.pkl', _refresh=True) +@cache_results('caches/weibo-lstm.pkl', _refresh=False) def get_data(): - data_bundle = ChineseNERLoader().process('MSRA-NER/', bigrams=True) - char_embed = StaticEmbedding(data_bundle.vocabs['chars'], - model_dir_or_name='cn-char') - bigram_embed = StaticEmbedding(data_bundle.vocabs['bigrams'], - model_dir_or_name='cn-bigram') + data_bundle = WeiboNERLoader().load() + data_bundle = ChineseNERPipe(encoding_type='bioes', bigram=True).process(data_bundle) + char_embed = StaticEmbedding(data_bundle.get_vocab(C.CHAR_INPUT), model_dir_or_name='cn-fasttext') + bigram_embed = StaticEmbedding(data_bundle.get_vocab('bigrams'), embedding_dim=100, min_freq=3) return data_bundle, char_embed, bigram_embed data_bundle, char_embed, bigram_embed = get_data() +# data_bundle = get_data() print(data_bundle) + # exit(0) -data_bundle.datasets['train'].set_input('target') -data_bundle.datasets['dev'].set_input('target') model = CNBiLSTMCRFNER(char_embed, num_classes=len(data_bundle.vocabs['target']), bigram_embed=bigram_embed) -Trainer(data_bundle.datasets['train'], model, batch_size=640, +Trainer(data_bundle.datasets['train'], model, batch_size=20, metrics=SpanFPreRecMetric(data_bundle.vocabs['target'], encoding_type='bioes'), - num_workers=2, dev_data=data_bundle. datasets['dev'], device=3).train() + num_workers=2, dev_data=data_bundle. datasets['dev'], device=0).train() diff --git a/reproduction/seqence_labelling/cws/data/CWSDataLoader.py b/reproduction/seqence_labelling/cws/data/CWSDataLoader.py deleted file mode 100644 index 3c82d814..00000000 --- a/reproduction/seqence_labelling/cws/data/CWSDataLoader.py +++ /dev/null @@ -1,249 +0,0 @@ - -from fastNLP.io.embed_loader import EmbeddingOption, EmbedLoader -from fastNLP.core.vocabulary import VocabularyOption -from fastNLP.io.base_loader import DataSetLoader, DataBundle -from typing import Union, Dict, List, Iterator -from fastNLP import DataSet -from fastNLP import Instance -from fastNLP import Vocabulary -from fastNLP import Const -from reproduction.utils import check_dataloader_paths -from functools import partial - -class SigHanLoader(DataSetLoader): - """ - 任务相关的说明可以在这里找到http://sighan.cs.uchicago.edu/ - 支持的数据格式为，一行一句，不同的word用空格隔开。如下例 - - 共同 创造 美好 的 新 世纪 —— 二○○一年 新年 - 女士 们 ， 先生 们 ， 同志 们 ， 朋友 们 ： - - 读取sighan中的数据集，返回的DataSet将包含以下的内容fields: - raw_chars: list(str), 每个元素是一个汉字 - chars: list(str), 每个元素是一个index(汉字对应的index) - target: list(int), 根据不同的encoding_type会有不同的变化 - - :param target_type: target的类型，当前支持以下的两种: "bmes", "shift_relay" - """ - - def __init__(self, target_type:str): - super().__init__() - - if target_type.lower() not in ('bmes', 'shift_relay'): - raise ValueError("target_type only supports 'bmes', 'shift_relay'.") - - self.target_type = target_type - if target_type=='bmes': - self._word_len_to_target = self._word_len_to_bems - elif target_type=='shift_relay': - self._word_len_to_target = self._word_lens_to_relay - - @staticmethod - def _word_lens_to_relay(word_lens: Iterator[int]): - """ - [1, 2, 3, ..] 转换为[0, 1, 0, 2, 1, 0,](start指示seg有多长); - :param word_lens: - :return: {'target': , 'end_seg_mask':, 'start_seg_mask':} - """ - tags = [] - end_seg_mask = [] - start_seg_mask = [] - for word_len in word_lens: - tags.extend([idx for idx in range(word_len - 1, -1, -1)]) - end_seg_mask.extend([0] * (word_len - 1) + [1]) - start_seg_mask.extend([1] + [0] * (word_len - 1)) - return {'target': tags, 'end_seg_mask': end_seg_mask, 'start_seg_mask': start_seg_mask} - - @staticmethod - def _word_len_to_bems(word_lens:Iterator[int])->Dict[str, List[str]]: - """ - - :param word_lens: 每个word的长度 - :return: - """ - tags = [] - for word_len in word_lens: - if word_len==1: - tags.append('S') - else: - tags.append('B') - for _ in range(word_len-2): - tags.append('M') - tags.append('E') - return {'target':tags} - - @staticmethod - def _gen_bigram(chars:List[str])->List[str]: - """ - - :param chars: - :return: - """ - return [c1+c2 for c1, c2 in zip(chars, chars[1:]+[''])] - - def load(self, path:str, bigram:bool=False)->DataSet: - """ - :param path: str - :param bigram: 是否使用bigram feature - :return: - """ - dataset = DataSet() - with open(path, 'r', encoding='utf-8') as f: - for line in f: - line = line.strip() - if not line: # 去掉空行 - continue - parts = line.split() - word_lens = map(len, parts) - chars = list(''.join(parts)) - tags = self._word_len_to_target(word_lens) - assert len(chars)==len(tags['target']) - dataset.append(Instance(raw_chars=chars, **tags, seq_len=len(chars))) - if len(dataset)==0: - raise RuntimeError(f"{path} has no valid data.") - if bigram: - dataset.apply_field(self._gen_bigram, field_name='raw_chars', new_field_name='bigrams') - return dataset - - def process(self, paths: Union[str, Dict[str, str]], char_vocab_opt:VocabularyOption=None, - char_embed_opt:EmbeddingOption=None, bigram_vocab_opt:VocabularyOption=None, - bigram_embed_opt:EmbeddingOption=None, L:int=4): - """ - 支持的数据格式为一行一个sample，并且用空格隔开不同的词语。例如 - - Option:: - - 共同 创造 美好 的 新 世纪 —— 二○○一年 新年 贺词 - （ 二○○○年 十二月 三十一日 ） （ 附 图片 1 张 ） - 女士 们 ， 先生 们 ， 同志 们 ， 朋友 们 ： - - paths支持两种格式，第一种是str，第二种是Dict[str, str]. - - Option:: - - # 1. str类型 - # 1.1 传入具体的文件路径 - data = SigHanLoader('bmes').process('/path/to/cws/data.txt') # 将读取data.txt的内容 - # 包含以下的内容data.vocabs['chars']:Vocabulary对象， - # data.vocabs['target']: Vocabulary对象，根据encoding_type可能会没有该值 - # data.embeddings['chars']: Embedding对象. 只有提供了预训练的词向量的路径才有该项 - # data.datasets['train']: DataSet对象 - # 包含的field有: - # raw_chars: list[str], 每个元素是一个汉字 - # chars: list[int], 每个元素是汉字对应的index - # target: list[int], 根据encoding_type有对应的变化 - # 1.2 传入一个目录, 里面必须包含train.txt文件 - data = SigHanLoader('bmes').process('path/to/cws/') #将尝试在该目录下读取 train.txt, test.txt以及dev.txt - # 包含以下的内容data.vocabs['chars']: Vocabulary对象 - # data.vocabs['target']:Vocabulary对象 - # data.embeddings['chars']: 仅在提供了预训练embedding路径的情况下，为Embedding对象; - # data.datasets['train']: DataSet对象 - # 包含的field有: - # raw_chars: list[str], 每个元素是一个汉字 - # chars: list[int], 每个元素是汉字对应的index - # target: list[int], 根据encoding_type有对应的变化 - # data.datasets['dev']: DataSet对象，如果文件夹下包含了dev.txt；内容与data.datasets['train']一样 - - # 2. dict类型, key是文件的名称，value是对应的读取路径. 必须包含'train'这个key - paths = {'train': '/path/to/train/train.txt', 'test':'/path/to/test/test.txt', 'dev':'/path/to/dev/dev.txt'} - data = SigHanLoader(paths).process(paths) - # 结果与传入目录时是一致的，但是可以传入多个数据集。data.datasets中的key将与这里传入的一致 - - :param paths: 支持传入目录，文件路径，以及dict。 - :param char_vocab_opt: 用于构建chars的vocabulary参数，默认为min_freq=2 - :param char_embed_opt: 用于读取chars的Embedding的参数，默认不读取pretrained的embedding - :param bigram_vocab_opt: 用于构建bigram的vocabulary参数，默认不使用bigram, 仅在指定该参数的情况下会带有bigrams这个field。 - 为List[int], 每个instance长度与chars一样, abcde的bigram为ab bc cd de e - :param bigram_embed_opt: 用于读取预训练bigram的参数，仅在传入bigram_vocab_opt有效 - :param L: 当target_type为shift_relay时传入的segment长度 - :return: - """ - # 推荐大家使用这个check_data_loader_paths进行paths的验证 - paths = check_dataloader_paths(paths) - datasets = {} - data = DataBundle() - bigram = bigram_vocab_opt is not None - for name, path in paths.items(): - dataset = self.load(path, bigram=bigram) - datasets[name] = dataset - input_fields = [] - target_fields = [] - # 创建vocab - char_vocab = Vocabulary(min_freq=2) if char_vocab_opt is None else Vocabulary(**char_vocab_opt) - char_vocab.from_dataset(datasets['train'], field_name='raw_chars') - char_vocab.index_dataset(*datasets.values(), field_name='raw_chars', new_field_name='chars') - data.vocabs[Const.CHAR_INPUT] = char_vocab - input_fields.extend([Const.CHAR_INPUT, Const.INPUT_LEN, Const.TARGET]) - target_fields.append(Const.TARGET) - # 创建target - if self.target_type == 'bmes': - target_vocab = Vocabulary(unknown=None, padding=None) - target_vocab.add_word_lst(['B']*4+['M']*3+['E']*2+['S']) - target_vocab.index_dataset(*datasets.values(), field_name='target') - data.vocabs[Const.TARGET] = target_vocab - if char_embed_opt is not None: - char_embed = EmbedLoader.load_with_vocab(**char_embed_opt, vocab=char_vocab) - data.embeddings['chars'] = char_embed - if bigram: - bigram_vocab = Vocabulary(**bigram_vocab_opt) - bigram_vocab.from_dataset(datasets['train'], field_name='bigrams') - bigram_vocab.index_dataset(*datasets.values(), field_name='bigrams') - data.vocabs['bigrams'] = bigram_vocab - if bigram_embed_opt is not None: - bigram_embed = EmbedLoader.load_with_vocab(**bigram_embed_opt, vocab=bigram_vocab) - data.embeddings['bigrams'] = bigram_embed - input_fields.append('bigrams') - if self.target_type == 'shift_relay': - func = partial(self._clip_target, L=L) - for name, dataset in datasets.items(): - res = dataset.apply_field(func, field_name='target') - relay_target = [res_i[0] for res_i in res] - relay_mask = [res_i[1] for res_i in res] - dataset.add_field('relay_target', relay_target, is_input=True, is_target=False, ignore_type=False) - dataset.add_field('relay_mask', relay_mask, is_input=True, is_target=False, ignore_type=False) - if self.target_type == 'shift_relay': - input_fields.extend(['end_seg_mask']) - target_fields.append('start_seg_mask') - # 将dataset加入DataInfo - for name, dataset in datasets.items(): - dataset.set_input(*input_fields) - dataset.set_target(*target_fields) - data.datasets[name] = dataset - - return data - - @staticmethod - def _clip_target(target:List[int], L:int): - """ - - 只有在target_type为shift_relay的使用 - :param target: List[int] - :param L: - :return: - """ - relay_target_i = [] - tmp = [] - for j in range(len(target) - 1): - tmp.append(target[j]) - if target[j] > target[j + 1]: - pass - else: - relay_target_i.extend([L - 1 if t >= L else t for t in tmp[::-1]]) - tmp = [] - # 处理未结束的部分 - if len(tmp) == 0: - relay_target_i.append(0) - else: - tmp.append(target[-1]) - relay_target_i.extend([L - 1 if t >= L else t for t in tmp[::-1]]) - relay_mask_i = [] - j = 0 - while j < len(target): - seg_len = target[j] + 1 - if target[j] < L: - relay_mask_i.extend([0] * (seg_len)) - else: - relay_mask_i.extend([1] * (seg_len - L) + [0] * L) - j = seg_len + j - return relay_target_i, relay_mask_i - diff --git a/reproduction/seqence_labelling/cws/data/cws_shift_pipe.py b/reproduction/seqence_labelling/cws/data/cws_shift_pipe.py new file mode 100644 index 00000000..0ae4064d --- /dev/null +++ b/reproduction/seqence_labelling/cws/data/cws_shift_pipe.py @@ -0,0 +1,202 @@ +from fastNLP.io.pipe import Pipe +from fastNLP.io import DataBundle +from fastNLP.io.loader import CWSLoader +from fastNLP import Const +from itertools import chain +from fastNLP.io.pipe.utils import _indexize +from functools import partial +from fastNLP.io.pipe.cws import _find_and_replace_alpha_spans, _find_and_replace_digit_spans + + +def _word_lens_to_relay(word_lens): + """ + [1, 2, 3, ..] 转换为[0, 1, 0, 2, 1, 0,](start指示seg有多长); + :param word_lens: + :return: + """ + tags = [] + for word_len in word_lens: + tags.extend([idx for idx in range(word_len - 1, -1, -1)]) + return tags + +def _word_lens_to_end_seg_mask(word_lens): + """ + [1, 2, 3, ..] 转换为[0, 1, 0, 2, 1, 0,](start指示seg有多长); + :param word_lens: + :return: + """ + end_seg_mask = [] + for word_len in word_lens: + end_seg_mask.extend([0] * (word_len - 1) + [1]) + return end_seg_mask + +def _word_lens_to_start_seg_mask(word_lens): + """ + [1, 2, 3, ..] 转换为[0, 1, 0, 2, 1, 0,](start指示seg有多长); + :param word_lens: + :return: + """ + start_seg_mask = [] + for word_len in word_lens: + start_seg_mask.extend([1] + [0] * (word_len - 1)) + return start_seg_mask + + +class CWSShiftRelayPipe(Pipe): + """ + + :param str,None dataset_name: 支持'pku', 'msra', 'cityu', 'as', None + :param int L: ShiftRelay模型的超参数 + :param bool replace_num_alpha: 是否将数字和字母用特殊字符替换。 + :param bool bigrams: 是否增加一列bigram. bigram的构成是['复', '旦', '大', '学', ...]->["复旦", "旦大", ...] + :param bool trigrams: 是否增加一列trigram. trigram的构成是 ['复', '旦', '大', '学', ...]->["复旦大", "旦大学", ...] + """ + def __init__(self, dataset_name=None, L=5, replace_num_alpha=True, bigrams=True): + self.dataset_name = dataset_name + self.bigrams = bigrams + self.replace_num_alpha = replace_num_alpha + self.L = L + + def _tokenize(self, data_bundle): + """ + 将data_bundle中的'chars'列切分成一个一个的word. + 例如输入是"共同 创造 美好.."->[[共, 同], [创, 造], [...], ] + + :param data_bundle: + :return: + """ + def split_word_into_chars(raw_chars): + words = raw_chars.split() + chars = [] + for word in words: + char = [] + subchar = [] + for c in word: + if c=='<': + subchar.append(c) + continue + if c=='>' and subchar[0]=='<': + char.append(''.join(subchar)) + subchar = [] + if subchar: + subchar.append(c) + else: + char.append(c) + char.extend(subchar) + chars.append(char) + return chars + + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(split_word_into_chars, field_name=Const.CHAR_INPUT, + new_field_name=Const.CHAR_INPUT) + return data_bundle + + def process(self, data_bundle: DataBundle) -> DataBundle: + """ + 可以处理的DataSet需要包含raw_words列 + + .. csv-table:: + :header: "raw_words" + + "上海 浦东 开发 与 法制 建设 同步" + "新华社 上海 二月 十日 电 （ 记者 谢金虎 、 张持坚 ）" + "..." + + :param data_bundle: + :return: + """ + data_bundle.copy_field(Const.RAW_WORD, Const.CHAR_INPUT) + + if self.replace_num_alpha: + data_bundle.apply_field(_find_and_replace_alpha_spans, Const.CHAR_INPUT, Const.CHAR_INPUT) + data_bundle.apply_field(_find_and_replace_digit_spans, Const.CHAR_INPUT, Const.CHAR_INPUT) + + self._tokenize(data_bundle) + input_field_names = [Const.CHAR_INPUT] + target_field_names = [] + + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars:_word_lens_to_relay(map(len, chars)), field_name=Const.CHAR_INPUT, + new_field_name=Const.TARGET) + dataset.apply_field(lambda chars:_word_lens_to_start_seg_mask(map(len, chars)), field_name=Const.CHAR_INPUT, + new_field_name='start_seg_mask') + dataset.apply_field(lambda chars:_word_lens_to_end_seg_mask(map(len, chars)), field_name=Const.CHAR_INPUT, + new_field_name='end_seg_mask') + dataset.apply_field(lambda chars:list(chain(*chars)), field_name=Const.CHAR_INPUT, + new_field_name=Const.CHAR_INPUT) + target_field_names.append('start_seg_mask') + input_field_names.append('end_seg_mask') + if self.bigrams: + for name, dataset in data_bundle.datasets.items(): + dataset.apply_field(lambda chars: [c1+c2 for c1, c2 in zip(chars, chars[1:]+[''])], + field_name=Const.CHAR_INPUT, new_field_name='bigrams') + input_field_names.append('bigrams') + + _indexize(data_bundle, ['chars', 'bigrams'], []) + + func = partial(_clip_target, L=self.L) + for name, dataset in data_bundle.datasets.items(): + res = dataset.apply_field(func, field_name='target') + relay_target = [res_i[0] for res_i in res] + relay_mask = [res_i[1] for res_i in res] + dataset.add_field('relay_target', relay_target, is_input=True, is_target=False, ignore_type=False) + dataset.add_field('relay_mask', relay_mask, is_input=True, is_target=False, ignore_type=False) + input_field_names.append('relay_target') + input_field_names.append('relay_mask') + + input_fields = [Const.TARGET, Const.INPUT_LEN] + input_field_names + target_fields = [Const.TARGET, Const.INPUT_LEN] + target_field_names + for name, dataset in data_bundle.datasets.items(): + dataset.add_seq_len(Const.CHAR_INPUT) + + data_bundle.set_input(*input_fields) + data_bundle.set_target(*target_fields) + + return data_bundle + + def process_from_file(self, paths=None) -> DataBundle: + """ + + :param str paths: + :return: + """ + if self.dataset_name is None and paths is None: + raise RuntimeError("You have to set `paths` when calling process_from_file() or `dataset_name `when initialization.") + if self.dataset_name is not None and paths is not None: + raise RuntimeError("You cannot specify `paths` and `dataset_name` simultaneously") + data_bundle = CWSLoader(self.dataset_name).load(paths) + return self.process(data_bundle) + +def _clip_target(target, L:int): + """ + + 只有在target_type为shift_relay的使用 + :param target: List[int] + :param L: + :return: + """ + relay_target_i = [] + tmp = [] + for j in range(len(target) - 1): + tmp.append(target[j]) + if target[j] > target[j + 1]: + pass + else: + relay_target_i.extend([L - 1 if t >= L else t for t in tmp[::-1]]) + tmp = [] + # 处理未结束的部分 + if len(tmp) == 0: + relay_target_i.append(0) + else: + tmp.append(target[-1]) + relay_target_i.extend([L - 1 if t >= L else t for t in tmp[::-1]]) + relay_mask_i = [] + j = 0 + while j < len(target): + seg_len = target[j] + 1 + if target[j] < L: + relay_mask_i.extend([0] * (seg_len)) + else: + relay_mask_i.extend([1] * (seg_len - L) + [0] * L) + j = seg_len + j + return relay_target_i, relay_mask_i diff --git a/reproduction/seqence_labelling/cws/model/bilstm_crf_cws.py b/reproduction/seqence_labelling/cws/model/bilstm_crf_cws.py new file mode 100644 index 00000000..4f87a81c --- /dev/null +++ b/reproduction/seqence_labelling/cws/model/bilstm_crf_cws.py @@ -0,0 +1,60 @@ + +import torch +from fastNLP.modules import LSTM +from fastNLP.modules import allowed_transitions, ConditionalRandomField +from fastNLP import seq_len_to_mask +from torch import nn +from fastNLP import Const +import torch.nn.functional as F + +class BiLSTMCRF(nn.Module): + def __init__(self, char_embed, hidden_size, num_layers, target_vocab=None, bigram_embed=None, trigram_embed=None, + dropout=0.5): + super().__init__() + + embed_size = char_embed.embed_size + self.char_embed = char_embed + if bigram_embed: + embed_size += bigram_embed.embed_size + self.bigram_embed = bigram_embed + if trigram_embed: + embed_size += trigram_embed.embed_size + self.trigram_embed = trigram_embed + + self.lstm = LSTM(embed_size, hidden_size=hidden_size//2, bidirectional=True, batch_first=True, + num_layers=num_layers) + self.dropout = nn.Dropout(p=dropout) + self.fc = nn.Linear(hidden_size, len(target_vocab)) + + transitions = None + if target_vocab: + transitions = allowed_transitions(target_vocab, include_start_end=True, encoding_type='bmes') + + self.crf = ConditionalRandomField(num_tags=len(target_vocab), allowed_transitions=transitions) + + def _forward(self, chars, bigrams, trigrams, seq_len, target=None): + chars = self.char_embed(chars) + if bigrams is not None: + bigrams = self.bigram_embed(bigrams) + chars = torch.cat([chars, bigrams], dim=-1) + if trigrams is not None: + trigrams = self.trigram_embed(trigrams) + chars = torch.cat([chars, trigrams], dim=-1) + + output, _ = self.lstm(chars, seq_len) + output = self.dropout(output) + output = self.fc(output) + output = F.log_softmax(output, dim=-1) + mask = seq_len_to_mask(seq_len) + if target is None: + pred, _ = self.crf.viterbi_decode(output, mask) + return {Const.OUTPUT:pred} + else: + loss = self.crf.forward(output, tags=target, mask=mask) + return {Const.LOSS:loss} + + def forward(self, chars, seq_len, target, bigrams=None, trigrams=None): + return self._forward(chars, bigrams, trigrams, seq_len, target) + + def predict(self, chars, seq_len, bigrams=None, trigrams=None): + return self._forward(chars, bigrams, trigrams, seq_len) \ No newline at end of file diff --git a/reproduction/seqence_labelling/cws/model/model.py b/reproduction/seqence_labelling/cws/model/bilstm_shift_relay.py similarity index 74% rename from reproduction/seqence_labelling/cws/model/model.py rename to reproduction/seqence_labelling/cws/model/bilstm_shift_relay.py index de945ac3..4ce1cc51 100644 --- a/reproduction/seqence_labelling/cws/model/model.py +++ b/reproduction/seqence_labelling/cws/model/bilstm_shift_relay.py @@ -1,7 +1,5 @@ from torch import nn import torch -from fastNLP.embeddings import Embedding -import numpy as np from reproduction.seqence_labelling.cws.model.module import FeatureFunMax, SemiCRFShiftRelay from fastNLP.modules import LSTM @@ -21,25 +19,21 @@ class ShiftRelayCWSModel(nn.Module): :param num_bigram_per_char: 每个character对应的bigram的数量 :param drop_p: Dropout的大小 """ - def __init__(self, char_embed:Embedding, bigram_embed:Embedding, hidden_size:int=400, num_layers:int=1, - L:int=6, num_bigram_per_char:int=1, drop_p:float=0.2): + def __init__(self, char_embed, bigram_embed, hidden_size:int=400, num_layers:int=1, L:int=6, drop_p:float=0.2): super().__init__() - self.char_embedding = Embedding(char_embed, dropout=drop_p) - self._pretrained_embed = False - if isinstance(char_embed, np.ndarray): - self._pretrained_embed = True - self.bigram_embedding = Embedding(bigram_embed, dropout=drop_p) - self.lstm = LSTM(100 * (num_bigram_per_char + 1), hidden_size // 2, num_layers=num_layers, bidirectional=True, + self.char_embedding = char_embed + self.bigram_embedding = bigram_embed + self.lstm = LSTM(char_embed.embed_size+bigram_embed.embed_size, hidden_size // 2, num_layers=num_layers, + bidirectional=True, batch_first=True) self.feature_fn = FeatureFunMax(hidden_size, L) self.semi_crf_relay = SemiCRFShiftRelay(L) self.feat_drop = nn.Dropout(drop_p) self.reset_param() - # self.feature_fn.reset_parameters() def reset_param(self): for name, param in self.named_parameters(): - if 'embedding' in name and self._pretrained_embed: + if 'embedding' in name: continue if 'bias_hh' in name: nn.init.constant_(param, 0) @@ -51,10 +45,8 @@ class ShiftRelayCWSModel(nn.Module): nn.init.xavier_uniform_(param) def get_feats(self, chars, bigrams, seq_len): - batch_size, max_len = chars.size() chars = self.char_embedding(chars) bigrams = self.bigram_embedding(bigrams) - bigrams = bigrams.view(bigrams.size(0), max_len, -1) chars = torch.cat([chars, bigrams], dim=-1) feats, _ = self.lstm(chars, seq_len) feats = self.feat_drop(feats) diff --git a/reproduction/seqence_labelling/cws/readme.md b/reproduction/seqence_labelling/cws/readme.md new file mode 100644 index 00000000..a25bb0ed --- /dev/null +++ b/reproduction/seqence_labelling/cws/readme.md @@ -0,0 +1,32 @@ +四个数据集的统计信息，最原始的数据可以从[http://sighan.cs.uchicago.edu/bakeoff2005/]()下载。 + +| pku | # of sents | # of tokens | +| ----- | ---------- | ----------- | +| train | 17173 | 1650222 | +| dev | 1881 | 176226 | +| test | 1944 | 172733 | +| total | 20998 | 1999181 | + + +| cityu | # of sents | # of tokens | +| ----- | ---------- | ----------- | +| train | 47696 | 2164907 | +| dev | 5323 | 238447 | +| test | 1492 | 67690 | +| total | 54511 | 2471044 | + + +| msra | # of sents | # of tokens | +| ----- | ---------- | ----------- | +| train | 78242 | 3644550 | +| dev | 8676 | 405919 | +| test | 3985 | 184355 | +| total | 90903 | 4234824 | + + +| as | # of sents | # of tokens | +| ----- | ---------- | ----------- | +| train | 638273 | 7536586 | +| dev | 70680 | 831464 | +| test | 14429 | 197681 | +| total | 723382 | 8565731 | diff --git a/reproduction/seqence_labelling/cws/test/__init__.py b/reproduction/seqence_labelling/cws/test/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/reproduction/seqence_labelling/cws/test/test_CWSDataLoader.py b/reproduction/seqence_labelling/cws/test/test_CWSDataLoader.py deleted file mode 100644 index f4260849..00000000 --- a/reproduction/seqence_labelling/cws/test/test_CWSDataLoader.py +++ /dev/null @@ -1,17 +0,0 @@ - - -import unittest -from ..data.CWSDataLoader import SigHanLoader -from fastNLP.core.vocabulary import VocabularyOption - - -class TestCWSDataLoader(unittest.TestCase): - def test_case1(self): - cws_loader = SigHanLoader(target_type='bmes') - data = cws_loader.process('pku_demo.txt') - print(data.datasets) - - def test_calse2(self): - cws_loader = SigHanLoader(target_type='bmes') - data = cws_loader.process('pku_demo.txt', bigram_vocab_opt=VocabularyOption()) - print(data.datasets) \ No newline at end of file diff --git a/reproduction/seqence_labelling/cws/train_bilstm_crf.py b/reproduction/seqence_labelling/cws/train_bilstm_crf.py new file mode 100644 index 00000000..b9a77249 --- /dev/null +++ b/reproduction/seqence_labelling/cws/train_bilstm_crf.py @@ -0,0 +1,52 @@ +import sys +sys.path.append('../../..') + +from fastNLP.io.pipe.cws import CWSPipe +from reproduction.seqence_labelling.cws.model.bilstm_crf_cws import BiLSTMCRF +from fastNLP import Trainer, cache_results +from fastNLP.embeddings import StaticEmbedding +from fastNLP import EvaluateCallback, BucketSampler, SpanFPreRecMetric, GradientClipCallback +from torch.optim import Adagrad + +###########hyper +dataname = 'pku' +hidden_size = 400 +num_layers = 1 +lr = 0.05 +###########hyper + + +@cache_results('{}.pkl'.format(dataname), _refresh=False) +def get_data(): + data_bundle = CWSPipe(dataset_name=dataname, bigrams=True, trigrams=False).process_from_file() + char_embed = StaticEmbedding(data_bundle.get_vocab('chars'), dropout=0.33, word_dropout=0.01, + model_dir_or_name='~/exps/CWS/pretrain/vectors/1grams_t3_m50_corpus.txt') + bigram_embed = StaticEmbedding(data_bundle.get_vocab('bigrams'), dropout=0.33,min_freq=3, word_dropout=0.01, + model_dir_or_name='~/exps/CWS/pretrain/vectors/2grams_t3_m50_corpus.txt') + return data_bundle, char_embed, bigram_embed + +data_bundle, char_embed, bigram_embed = get_data() +print(data_bundle) + +model = BiLSTMCRF(char_embed, hidden_size, num_layers, target_vocab=data_bundle.get_vocab('target'), bigram_embed=bigram_embed, + trigram_embed=None, dropout=0.3) +model.cuda() + +callbacks = [] +callbacks.append(EvaluateCallback(data_bundle.get_dataset('test'))) +callbacks.append(GradientClipCallback(clip_type='value', clip_value=5)) +optimizer = Adagrad(model.parameters(), lr=lr) + +metrics = [] +metric1 = SpanFPreRecMetric(tag_vocab=data_bundle.get_vocab('target'), encoding_type='bmes') +metrics.append(metric1) + +trainer = Trainer(data_bundle.get_dataset('train'), model, optimizer=optimizer, loss=None, + batch_size=128, sampler=BucketSampler(), update_every=1, + num_workers=1, n_epochs=10, print_every=5, + dev_data=data_bundle.get_dataset('dev'), + metrics=metrics, + metric_key=None, + validate_every=-1, save_path=None, use_tqdm=True, device=0, + callbacks=callbacks, check_code_level=0, dev_batch_size=128) +trainer.train() diff --git a/reproduction/seqence_labelling/cws/train_shift_relay.py b/reproduction/seqence_labelling/cws/train_shift_relay.py index 55576575..322f42bb 100644 --- a/reproduction/seqence_labelling/cws/train_shift_relay.py +++ b/reproduction/seqence_labelling/cws/train_shift_relay.py @@ -1,64 +1,53 @@ -import os +import sys +sys.path.append('../../..') from fastNLP import cache_results -from reproduction.seqence_labelling.cws.data.CWSDataLoader import SigHanLoader -from reproduction.seqence_labelling.cws.model.model import ShiftRelayCWSModel -from fastNLP.io.embed_loader import EmbeddingOption -from fastNLP.core.vocabulary import VocabularyOption +from reproduction.seqence_labelling.cws.data.cws_shift_pipe import CWSShiftRelayPipe +from reproduction.seqence_labelling.cws.model.bilstm_shift_relay import ShiftRelayCWSModel from fastNLP import Trainer from torch.optim import Adam from fastNLP import BucketSampler from fastNLP import GradientClipCallback from reproduction.seqence_labelling.cws.model.metric import RelayMetric - - -# 借助一下fastNLP的自动缓存机制，但是只能缓存4G以下的结果 -@cache_results(None) -def prepare_data(): - data = SigHanLoader(target_type='shift_relay').process(file_dir, char_embed_opt=char_embed_opt, - bigram_vocab_opt=bigram_vocab_opt, - bigram_embed_opt=bigram_embed_opt, - L=L) - return data +from fastNLP.embeddings import StaticEmbedding +from fastNLP import EvaluateCallback #########hyper L = 4 hidden_size = 200 num_layers = 1 drop_p = 0.2 -lr = 0.02 - +lr = 0.008 +data_name = 'pku' #########hyper device = 0 -# !!!!这里千万不要放完全路径，因为这样会暴露你们在服务器上的用户名，比较危险。所以一定要使用相对路径，最好把数据放到 -# 你们的reproduction路径下，然后设置.gitignore -file_dir = '/path/to/' -char_embed_path = '/pretrain/vectors/1grams_t3_m50_corpus.txt' -bigram_embed_path = '/pretrain/vectors/2grams_t3_m50_corpus.txt' -bigram_vocab_opt = VocabularyOption(min_freq=3) -char_embed_opt = EmbeddingOption(embed_filepath=char_embed_path) -bigram_embed_opt = EmbeddingOption(embed_filepath=bigram_embed_path) - -data_name = os.path.basename(file_dir) cache_fp = 'caches/{}.pkl'.format(data_name) +@cache_results(_cache_fp=cache_fp, _refresh=True) # 将结果缓存到cache_fp中，这样下次运行就直接读取，而不需要再次运行 +def prepare_data(): + data_bundle = CWSShiftRelayPipe(dataset_name=data_name, L=L).process_from_file() + # 预训练的character embedding和bigram embedding + char_embed = StaticEmbedding(data_bundle.get_vocab('chars'), dropout=0.5, word_dropout=0.01, + model_dir_or_name='~/exps/CWS/pretrain/vectors/1grams_t3_m50_corpus.txt') + bigram_embed = StaticEmbedding(data_bundle.get_vocab('bigrams'), dropout=0.5, min_freq=3, word_dropout=0.01, + model_dir_or_name='~/exps/CWS/pretrain/vectors/2grams_t3_m50_corpus.txt') -data = prepare_data(_cache_fp=cache_fp, _refresh=True) + return data_bundle, char_embed, bigram_embed -model = ShiftRelayCWSModel(char_embed=data.embeddings['chars'], bigram_embed=data.embeddings['bigrams'], - hidden_size=hidden_size, num_layers=num_layers, - L=L, num_bigram_per_char=1, drop_p=drop_p) +data, char_embed, bigram_embed = prepare_data() -sampler = BucketSampler(batch_size=32) +model = ShiftRelayCWSModel(char_embed=char_embed, bigram_embed=bigram_embed, + hidden_size=hidden_size, num_layers=num_layers, drop_p=drop_p, L=L) + +sampler = BucketSampler() optimizer = Adam(model.parameters(), lr=lr) -clipper = GradientClipCallback(clip_value=5, clip_type='value') -callbacks = [clipper] -# if pretrain: -# fixer = FixEmbedding([model.char_embedding, model.bigram_embedding], fix_until=fix_until) -# callbacks.append(fixer) -trainer = Trainer(data.datasets['train'], model, optimizer=optimizer, loss=None, batch_size=32, sampler=sampler, - update_every=5, n_epochs=3, print_every=5, dev_data=data.datasets['dev'], metrics=RelayMetric(), +clipper = GradientClipCallback(clip_value=5, clip_type='value') # 截断太大的梯度 +evaluator = EvaluateCallback(data.get_dataset('test')) # 额外测试在test集上的效果 +callbacks = [clipper, evaluator] + +trainer = Trainer(data.get_dataset('train'), model, optimizer=optimizer, loss=None, batch_size=128, sampler=sampler, + update_every=1, n_epochs=10, print_every=5, dev_data=data.get_dataset('dev'), metrics=RelayMetric(), metric_key='f', validate_every=-1, save_path=None, use_tqdm=True, device=device, callbacks=callbacks, - check_code_level=0) + check_code_level=0, num_workers=1) trainer.train() \ No newline at end of file diff --git a/reproduction/seqence_labelling/ner/data/Conll2003Loader.py b/reproduction/seqence_labelling/ner/data/Conll2003Loader.py deleted file mode 100644 index 1aeddcf8..00000000 --- a/reproduction/seqence_labelling/ner/data/Conll2003Loader.py +++ /dev/null @@ -1,93 +0,0 @@ - -from fastNLP.core.vocabulary import VocabularyOption -from fastNLP.io.base_loader import DataSetLoader, DataBundle -from typing import Union, Dict -from fastNLP import Vocabulary -from fastNLP import Const -from reproduction.utils import check_dataloader_paths - -from fastNLP.io import ConllLoader -from reproduction.seqence_labelling.ner.data.utils import iob2bioes, iob2 - - -class Conll2003DataLoader(DataSetLoader): - def __init__(self, task:str='ner', encoding_type:str='bioes'): - """ - 加载Conll2003格式的英语语料，该数据集的信息可以在https://www.clips.uantwerpen.be/conll2003/ner/找到。当task为pos - 时，返回的DataSet中target取值于第2列; 当task为chunk时，返回的DataSet中target取值于第3列;当task为ner时，返回 - 的DataSet中target取值于第4列。所有"-DOCSTART- -X- O O"将被忽略，这会导致数据的数量少于很多文献报道的值，但 - 鉴于"-DOCSTART- -X- O O"只是用于文档分割的符号，并不应该作为预测对象，所以我们忽略了数据中的-DOCTSTART-开头的行 - ner与chunk任务读取后的数据的target将为encoding_type类型。pos任务读取后就是pos列的数据。 - - :param task: 指定需要标注任务。可选ner, pos, chunk - """ - assert task in ('ner', 'pos', 'chunk') - index = {'ner':3, 'pos':1, 'chunk':2}[task] - self._loader = ConllLoader(headers=['raw_words', 'target'], indexes=[0, index]) - self._tag_converters = [] - if task in ('ner', 'chunk'): - self._tag_converters = [iob2] - if encoding_type == 'bioes': - self._tag_converters.append(iob2bioes) - - def load(self, path: str): - dataset = self._loader.load(path) - def convert_tag_schema(tags): - for converter in self._tag_converters: - tags = converter(tags) - return tags - if self._tag_converters: - dataset.apply_field(convert_tag_schema, field_name=Const.TARGET, new_field_name=Const.TARGET) - return dataset - - def process(self, paths: Union[str, Dict[str, str]], word_vocab_opt:VocabularyOption=None, lower:bool=False): - """ - 读取并处理数据。数据中的'-DOCSTART-'开头的行会被忽略 - - :param paths: - :param word_vocab_opt: vocabulary的初始化值 - :param lower: 是否将所有字母转为小写。 - :return: - """ - # 读取数据 - paths = check_dataloader_paths(paths) - data = DataBundle() - input_fields = [Const.TARGET, Const.INPUT, Const.INPUT_LEN] - target_fields = [Const.TARGET, Const.INPUT_LEN] - for name, path in paths.items(): - dataset = self.load(path) - dataset.apply_field(lambda words: words, field_name='raw_words', new_field_name=Const.INPUT) - if lower: - dataset.words.lower() - data.datasets[name] = dataset - - # 对construct vocab - word_vocab = Vocabulary(min_freq=2) if word_vocab_opt is None else Vocabulary(**word_vocab_opt) - word_vocab.from_dataset(data.datasets['train'], field_name=Const.INPUT, - no_create_entry_dataset=[dataset for name, dataset in data.datasets.items() if name!='train']) - word_vocab.index_dataset(*data.datasets.values(), field_name=Const.INPUT, new_field_name=Const.INPUT) - data.vocabs[Const.INPUT] = word_vocab - - # cap words - cap_word_vocab = Vocabulary() - cap_word_vocab.from_dataset(data.datasets['train'], field_name='raw_words', - no_create_entry_dataset=[dataset for name, dataset in data.datasets.items() if name!='train']) - cap_word_vocab.index_dataset(*data.datasets.values(), field_name='raw_words', new_field_name='cap_words') - input_fields.append('cap_words') - data.vocabs['cap_words'] = cap_word_vocab - - # 对target建vocab - target_vocab = Vocabulary(unknown=None, padding=None) - target_vocab.from_dataset(*data.datasets.values(), field_name=Const.TARGET) - target_vocab.index_dataset(*data.datasets.values(), field_name=Const.TARGET) - data.vocabs[Const.TARGET] = target_vocab - - for name, dataset in data.datasets.items(): - dataset.add_seq_len(Const.INPUT, new_field_name=Const.INPUT_LEN) - dataset.set_input(*input_fields) - dataset.set_target(*target_fields) - - return data - -if __name__ == '__main__': - pass \ No newline at end of file diff --git a/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py b/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py deleted file mode 100644 index a6070f39..00000000 --- a/reproduction/seqence_labelling/ner/data/OntoNoteLoader.py +++ /dev/null @@ -1,152 +0,0 @@ -from fastNLP.core.vocabulary import VocabularyOption -from fastNLP.io.base_loader import DataSetLoader, DataBundle -from typing import Union, Dict -from fastNLP import DataSet -from fastNLP import Vocabulary -from fastNLP import Const -from reproduction.utils import check_dataloader_paths - -from fastNLP.io import ConllLoader -from reproduction.seqence_labelling.ner.data.utils import iob2bioes, iob2 - -class OntoNoteNERDataLoader(DataSetLoader): - """ - 用于读取处理为Conll格式后的OntoNote数据。将OntoNote数据处理为conll格式的过程可以参考https://github.com/yhcc/OntoNotes-5.0-NER。 - - """ - def __init__(self, encoding_type:str='bioes'): - assert encoding_type in ('bioes', 'bio') - self.encoding_type = encoding_type - if encoding_type=='bioes': - self.encoding_method = iob2bioes - else: - self.encoding_method = iob2 - - def load(self, path:str)->DataSet: - """ - 给定一个文件路径，读取数据。返回的DataSet包含以下的field - raw_words: List[str] - target: List[str] - - :param path: - :return: - """ - dataset = ConllLoader(headers=['raw_words', 'target'], indexes=[3, 10]).load(path) - def convert_to_bio(tags): - bio_tags = [] - flag = None - for tag in tags: - label = tag.strip("()*") - if '(' in tag: - bio_label = 'B-' + label - flag = label - elif flag: - bio_label = 'I-' + flag - else: - bio_label = 'O' - if ')' in tag: - flag = None - bio_tags.append(bio_label) - return self.encoding_method(bio_tags) - - def convert_word(words): - converted_words = [] - for word in words: - word = word.replace('/.', '.') # 有些结尾的.是/.形式的 - if not word.startswith('-'): - converted_words.append(word) - continue - # 以下是由于这些符号被转义了，再转回来 - tfrs = {'-LRB-':'(', - '-RRB-': ')', - '-LSB-': '[', - '-RSB-': ']', - '-LCB-': '{', - '-RCB-': '}' - } - if word in tfrs: - converted_words.append(tfrs[word]) - else: - converted_words.append(word) - return converted_words - - dataset.apply_field(convert_word, field_name='raw_words', new_field_name='raw_words') - dataset.apply_field(convert_to_bio, field_name='target', new_field_name='target') - - return dataset - - def process(self, paths: Union[str, Dict[str, str]], word_vocab_opt:VocabularyOption=None, - lower:bool=True)->DataBundle: - """ - 读取并处理数据。返回的DataInfo包含以下的内容 - vocabs: - word: Vocabulary - target: Vocabulary - datasets: - train: DataSet - words: List[int], 被设置为input - target: int. label，被同时设置为input和target - seq_len: int. 句子的长度，被同时设置为input和target - raw_words: List[str] - xxx(根据传入的paths可能有所变化) - - :param paths: - :param word_vocab_opt: vocabulary的初始化值 - :param lower: 是否使用小写 - :return: - """ - paths = check_dataloader_paths(paths) - data = DataBundle() - input_fields = [Const.TARGET, Const.INPUT, Const.INPUT_LEN] - target_fields = [Const.TARGET, Const.INPUT_LEN] - for name, path in paths.items(): - dataset = self.load(path) - dataset.apply_field(lambda words: words, field_name='raw_words', new_field_name=Const.INPUT) - if lower: - dataset.words.lower() - data.datasets[name] = dataset - - # 对construct vocab - word_vocab = Vocabulary(min_freq=2) if word_vocab_opt is None else Vocabulary(**word_vocab_opt) - word_vocab.from_dataset(data.datasets['train'], field_name=Const.INPUT, - no_create_entry_dataset=[dataset for name, dataset in data.datasets.items() if name!='train']) - word_vocab.index_dataset(*data.datasets.values(), field_name=Const.INPUT, new_field_name=Const.INPUT) - data.vocabs[Const.INPUT] = word_vocab - - # cap words - cap_word_vocab = Vocabulary() - cap_word_vocab.from_dataset(*data.datasets.values(), field_name='raw_words') - cap_word_vocab.index_dataset(*data.datasets.values(), field_name='raw_words', new_field_name='cap_words') - input_fields.append('cap_words') - data.vocabs['cap_words'] = cap_word_vocab - - # 对target建vocab - target_vocab = Vocabulary(unknown=None, padding=None) - target_vocab.from_dataset(*data.datasets.values(), field_name=Const.TARGET) - target_vocab.index_dataset(*data.datasets.values(), field_name=Const.TARGET) - data.vocabs[Const.TARGET] = target_vocab - - for name, dataset in data.datasets.items(): - dataset.add_seq_len(Const.INPUT, new_field_name=Const.INPUT_LEN) - dataset.set_input(*input_fields) - dataset.set_target(*target_fields) - - return data - - -if __name__ == '__main__': - loader = OntoNoteNERDataLoader() - dataset = loader.load('/hdd/fudanNLP/fastNLP/others/data/v4/english/test.txt') - print(dataset.target.value_count()) - print(dataset[:4]) - - -""" -train 115812 2200752 -development 15680 304684 -test 12217 230111 - -train 92403 1901772 -valid 13606 279180 -test 10258 204135 -""" \ No newline at end of file diff --git a/reproduction/seqence_labelling/ner/data/utils.py b/reproduction/seqence_labelling/ner/data/utils.py deleted file mode 100644 index 8f7af792..00000000 --- a/reproduction/seqence_labelling/ner/data/utils.py +++ /dev/null @@ -1,49 +0,0 @@ -from typing import List - -def iob2(tags:List[str])->List[str]: - """ - 检查数据是否是合法的IOB数据，如果是IOB1会被自动转换为IOB2。 - - :param tags: 需要转换的tags - """ - for i, tag in enumerate(tags): - if tag == "O": - continue - split = tag.split("-") - if len(split) != 2 or split[0] not in ["I", "B"]: - raise TypeError("The encoding schema is not a valid IOB type.") - if split[0] == "B": - continue - elif i == 0 or tags[i - 1] == "O": # conversion IOB1 to IOB2 - tags[i] = "B" + tag[1:] - elif tags[i - 1][1:] == tag[1:]: - continue - else: # conversion IOB1 to IOB2 - tags[i] = "B" + tag[1:] - return tags - -def iob2bioes(tags:List[str])->List[str]: - """ - 将iob的tag转换为bmeso编码 - :param tags: - :return: - """ - new_tags = [] - for i, tag in enumerate(tags): - if tag == 'O': - new_tags.append(tag) - else: - split = tag.split('-')[0] - if split == 'B': - if i+1!=len(tags) and tags[i+1].split('-')[0] == 'I': - new_tags.append(tag) - else: - new_tags.append(tag.replace('B-', 'S-')) - elif split == 'I': - if i + 1=3.4.1 prettytable>=0.7.2 requests spacy +prettytable>=0.7.2 \ No newline at end of file diff --git a/test/core/test_dataset.py b/test/core/test_dataset.py index 0228f207..9820eff6 100644 --- a/test/core/test_dataset.py +++ b/test/core/test_dataset.py @@ -1,4 +1,5 @@ import os +import sys import unittest from fastNLP import DataSet @@ -79,6 +80,16 @@ class TestDataSetMethods(unittest.TestCase): self.assertFalse("x" in dd.field_arrays) self.assertTrue("y" in dd.field_arrays) + def test_delete_instance(self): + dd = DataSet() + old_length = 2 + dd.add_field("x", [[1, 2, 3]] * old_length) + dd.add_field("y", [[1, 2, 3, 4]] * old_length) + dd.delete_instance(0) + self.assertEqual(len(dd), old_length-1) + dd.delete_instance(0) + self.assertEqual(len(dd), old_length-2) + def test_getitem(self): ds = DataSet({"x": [[1, 2, 3, 4]] * 40, "y": [[5, 6]] * 40}) ins_1, ins_0 = ds[0], ds[1] @@ -171,8 +182,9 @@ class TestDataSetMethods(unittest.TestCase): def test_apply2(self): def split_sent(ins): return ins['raw_sentence'].split() - csv_loader = CSVLoader(headers=['raw_sentence', 'label'],sep='\t') - dataset = csv_loader.load('test/data_for_tests/tutorial_sample_dataset.csv') + csv_loader = CSVLoader(headers=['raw_sentence', 'label'], sep='\t') + data_bundle = csv_loader.load('test/data_for_tests/tutorial_sample_dataset.csv') + dataset = data_bundle.datasets['train'] dataset.drop(lambda x: len(x['raw_sentence'].split()) == 0, inplace=True) dataset.apply(split_sent, new_field_name='words', is_input=True) # print(dataset) @@ -217,4 +229,17 @@ class TestDataSetIter(unittest.TestCase): def test__repr__(self): ds = DataSet({"x": [[1, 2, 3, 4]] * 10, "y": [[5, 6]] * 10}) for iter in ds: - self.assertEqual(iter.__repr__(), "{'x': [1, 2, 3, 4] type=list,\n'y': [5, 6] type=list}") + self.assertEqual(iter.__repr__(), """+--------------+--------+ +| x | y | ++--------------+--------+ +| [1, 2, 3, 4] | [5, 6] | ++--------------+--------+""") + + +class TestDataSetFieldMeta(unittest.TestCase): + def test_print_field_meta(self): + ds = DataSet({"x": [[1, 2, 3, 4]] * 10, "y": [[5, 6]] * 10}) + ds.print_field_meta() + + ds.set_input('x') + ds.print_field_meta() diff --git a/test/core/test_dist_trainer.py b/test/core/test_dist_trainer.py new file mode 100644 index 00000000..c6879634 --- /dev/null +++ b/test/core/test_dist_trainer.py @@ -0,0 +1,167 @@ +import unittest + +import numpy as np +import torch.cuda +from fastNLP import DataSet +from fastNLP import Instance +from fastNLP import CrossEntropyLoss, BCELoss +from fastNLP import SGD +from fastNLP.core.dist_trainer import DistTrainer, get_local_rank +from fastNLP.models.base_model import NaiveClassifier +import shutil +import os +import subprocess +from argparse import ArgumentParser +from fastNLP.core.callback import EchoCallback +from fastNLP import AccuracyMetric + +def prepare_fake_dataset(): + mean = np.array([-3, -3]) + cov = np.array([[1, 0], [0, 1]]) + class_A = np.random.multivariate_normal(mean, cov, size=(1000,)) + + mean = np.array([3, 3]) + cov = np.array([[1, 0], [0, 1]]) + class_B = np.random.multivariate_normal(mean, cov, size=(1000,)) + + data_set = DataSet([Instance(x=[float(item[0]), float(item[1])], y=0) for item in class_A] + + [Instance(x=[float(item[0]), float(item[1])], y=1) for item in class_B]) + return data_set + +def prepare_fake_dataset2(*args, size=100): + ys = np.random.randint(4, size=100, dtype=np.int64) + data = {'y': ys} + for arg in args: + data[arg] = np.random.randn(size, 5) + return DataSet(data=data) + +def set_rng_seed(seed): + np.random.seed(seed) + +def prepare_env(): + def prepare_fake_dataset(): + mean = np.array([-3, -3]) + cov = np.array([[1, 0], [0, 1]]) + class_A = np.random.multivariate_normal(mean, cov, size=(1000,)) + + mean = np.array([3, 3]) + cov = np.array([[1, 0], [0, 1]]) + class_B = np.random.multivariate_normal(mean, cov, size=(1000,)) + + data_set = DataSet([Instance(x=[float(item[0]), float(item[1])], y=[0.0]) for item in class_A] + + [Instance(x=[float(item[0]), float(item[1])], y=[1.0]) for item in class_B]) + return data_set + + data_set = prepare_fake_dataset() + data_set.set_input("x") + data_set.set_target("y") + model = NaiveClassifier(2, 1) + return data_set, model + +class TestDistTrainer(unittest.TestCase): + save_path = './save_cp' + + def run1(self): + # test distributed training + print('local rank', get_local_rank()) + set_rng_seed(100) + data_set = prepare_fake_dataset() + data_set.set_input("x", flag=True) + data_set.set_target("y", flag=True) + + model = NaiveClassifier(2, 2) + + trainer = DistTrainer( + model=model, train_data=data_set, optimizer=SGD(lr=0.1), + loss=CrossEntropyLoss(pred="predict", target="y"), + batch_size_per_gpu=8, n_epochs=3, print_every=50, save_path=self.save_path, + ) + trainer.train() + """ + # 应该正确运行 + """ + if trainer.is_master and os.path.exists(self.save_path): + shutil.rmtree(self.save_path) + + def run2(self): + # test fp16 with distributed training + print('local rank', get_local_rank()) + set_rng_seed(100) + data_set = prepare_fake_dataset() + data_set.set_input("x", flag=True) + data_set.set_target("y", flag=True) + + model = NaiveClassifier(2, 2) + + trainer = DistTrainer( + model=model, train_data=data_set, optimizer=SGD(lr=0.1), + loss=CrossEntropyLoss(pred="predict", target="y"), + batch_size_per_gpu=8, n_epochs=3, print_every=50, save_path=self.save_path, + fp16='O1' + ) + trainer.train() + """ + # 应该正确运行 + """ + if trainer.is_master and os.path.exists(self.save_path): + shutil.rmtree(self.save_path) + + def run3(self): + set_rng_seed(100) + data_set, model = prepare_env() + trainer = DistTrainer( + data_set, model, optimizer=None, + loss=BCELoss(pred="predict", target="y"), + n_epochs=3, print_every=50, + callbacks_all=[EchoCallback('callbacks_all')], + callbacks_master=[EchoCallback('callbacks_master')] + ) + trainer.train() + + def run4(self): + set_rng_seed(100) + data_set, model = prepare_env() + + train_set, dev_set = data_set.split(0.3) + + model = NaiveClassifier(2, 1) + + trainer = DistTrainer( + train_set, model, optimizer=SGD(lr=0.1), + loss=BCELoss(pred="predict", target="y"), + batch_size_per_gpu=32, n_epochs=3, print_every=50, dev_data=dev_set, + metrics=AccuracyMetric(pred="predict", target="y"), validate_every=-1, save_path=None, + ) + trainer.train() + """ + # 应该正确运行 + """ + + def run_dist(self, run_id): + if torch.cuda.is_available(): + ngpu = min(2, torch.cuda.device_count()) + path = __file__ + cmd = ['python', '-m', 'torch.distributed.launch', + '--nproc_per_node', str(ngpu), path, '--test', str(run_id)] + print(' '.join(cmd)) + subprocess.check_call(cmd) + + def test_normal_run(self): + self.run_dist(1) + + def no_test_fp16(self): + self.run_dist(2) + + def test_callback(self): + self.run_dist(3) + + def test_dev_data(self): + self.run_dist(4) + +if __name__ == '__main__': + runner = TestDistTrainer() + parser = ArgumentParser() + parser.add_argument('--test', type=int) + args, _ = parser.parse_known_args() + if args.test and hasattr(runner, 'run%s'%args.test): + getattr(runner, 'run%s'%args.test)() diff --git a/test/core/test_field.py b/test/core/test_field.py index e9053f37..c46e2de2 100644 --- a/test/core/test_field.py +++ b/test/core/test_field.py @@ -170,22 +170,22 @@ class TestFieldArray(unittest.TestCase): def test_append(self): with self.assertRaises(Exception): - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append(0) with self.assertRaises(Exception): - fa = FieldArray("y", [1.1, 2.2, 3.3, 4.4, 5.5], is_input=True) + fa = FieldArray("y", [1.1, 2.2, 3.3, 4.4, 5.5], is_input=True, use_1st_ins_infer_dim_type=False) fa.append([1, 2, 3, 4, 5]) with self.assertRaises(Exception): - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append([]) with self.assertRaises(Exception): - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1, 2, 3, 4, 5]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append(["str", 0, 0, 0, 1.89]) - fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1.0, 2.0, 3.0, 4.0, 5.0]], is_input=True) + fa = FieldArray("y", [[1.1, 2.2, 3.3, 4.4, 5.5], [1.0, 2.0, 3.0, 4.0, 5.0]], is_input=True, use_1st_ins_infer_dim_type=False) fa.append([1.2, 2.3, 3.4, 4.5, 5.6]) self.assertEqual(len(fa), 3) self.assertEqual(fa[2], [1.2, 2.3, 3.4, 4.5, 5.6]) diff --git a/test/core/test_metrics.py b/test/core/test_metrics.py index 9c8a586c..8a472a62 100644 --- a/test/core/test_metrics.py +++ b/test/core/test_metrics.py @@ -7,10 +7,16 @@ from fastNLP import AccuracyMetric from fastNLP.core.metrics import _pred_topk, _accuracy_topk from fastNLP.core.vocabulary import Vocabulary from collections import Counter -from fastNLP.core.metrics import SpanFPreRecMetric +from fastNLP.core.metrics import SpanFPreRecMetric, ExtractiveQAMetric def _generate_tags(encoding_type, number_labels=4): + """ + + :param encoding_type: 例如BIOES, BMES, BIO等 + :param number_labels: 多少个label，大于1 + :return: + """ vocab = {} for i in range(number_labels): label = str(i) @@ -184,7 +190,7 @@ class TestAccuracyMetric(unittest.TestCase): self.assertDictEqual(metric.get_metric(), {'acc': 1.}) -class SpanF1PreRecMetric(unittest.TestCase): +class SpanFPreRecMetricTest(unittest.TestCase): def test_case1(self): from fastNLP.core.metrics import _bmes_tag_to_spans from fastNLP.core.metrics import _bio_tag_to_spans @@ -338,6 +344,74 @@ class SpanF1PreRecMetric(unittest.TestCase): for key, value in expected_metric.items(): self.assertAlmostEqual(value, metric_value[key], places=5) + def test_auto_encoding_type_infer(self): + # 检查是否可以自动check encode的类型 + vocabs = {} + import random + for encoding_type in ['bio', 'bioes', 'bmeso']: + vocab = Vocabulary(unknown=None, padding=None) + for i in range(random.randint(10, 100)): + label = str(random.randint(1, 10)) + for tag in encoding_type: + if tag!='o': + vocab.add_word(f'{tag}-{label}') + else: + vocab.add_word('o') + vocabs[encoding_type] = vocab + for e in ['bio', 'bioes', 'bmeso']: + with self.subTest(e=e): + metric = SpanFPreRecMetric(tag_vocab=vocabs[e]) + assert metric.encoding_type == e + + bmes_vocab = _generate_tags('bmes') + vocab = Vocabulary() + for tag, index in bmes_vocab.items(): + vocab.add_word(tag) + metric = SpanFPreRecMetric(vocab) + assert metric.encoding_type == 'bmes' + + # 一些无法check的情况 + vocab = Vocabulary() + for i in range(10): + vocab.add_word(str(i)) + with self.assertRaises(Exception): + metric = SpanFPreRecMetric(vocab) + + def test_encoding_type(self): + # 检查传入的tag_vocab与encoding_type不符合时，是否会报错 + vocabs = {} + import random + from itertools import product + for encoding_type in ['bio', 'bioes', 'bmeso']: + vocab = Vocabulary(unknown=None, padding=None) + for i in range(random.randint(10, 100)): + label = str(random.randint(1, 10)) + for tag in encoding_type: + if tag!='o': + vocab.add_word(f'{tag}-{label}') + else: + vocab.add_word('o') + vocabs[encoding_type] = vocab + for e1, e2 in product(['bio', 'bioes', 'bmeso'], ['bio', 'bioes', 'bmeso']): + with self.subTest(e1=e1, e2=e2): + if e1==e2: + metric = SpanFPreRecMetric(vocabs[e1], encoding_type=e2) + else: + s2 = set(e2) + s2.update(set(e1)) + if s2==set(e2): + continue + with self.assertRaises(AssertionError): + metric = SpanFPreRecMetric(vocabs[e1], encoding_type=e2) + for encoding_type in ['bio', 'bioes', 'bmeso']: + with self.assertRaises(AssertionError): + metric = SpanFPreRecMetric(vocabs[encoding_type], encoding_type='bmes') + + with self.assertWarns(Warning): + vocab = Vocabulary(unknown=None, padding=None).add_word_lst(list('bmes')) + metric = SpanFPreRecMetric(vocab, encoding_type='bmeso') + vocab = Vocabulary().add_word_lst(list('bmes')) + metric = SpanFPreRecMetric(vocab, encoding_type='bmeso') class TestUsefulFunctions(unittest.TestCase): # 测试metrics.py中一些看上去挺有用的函数 @@ -347,3 +421,46 @@ class TestUsefulFunctions(unittest.TestCase): _ = _pred_topk(np.random.randint(0, 3, size=(10, 1))) # 跑通即可 + + +class TestExtractiveQAMetric(unittest.TestCase): + + def test_cast_1(self): + qa_prediction = torch.FloatTensor([[[-0.4424, -0.4579, -0.7376, 1.8129, 0.1316, 1.6566, -1.2169, + -0.3782, 0.8240], + [-1.2348, -0.1876, -0.1462, -0.4834, -0.6692, -0.9735, -1.1563, + -0.3562, -1.4116], + [-1.6550, -0.9555, 0.3782, -1.3160, -1.5835, -0.3443, -1.7858, + -2.0023, 0.0075], + [-0.3772, -0.5447, -1.5631, 1.1614, 1.4598, -1.2764, 0.5186, + 0.3832, -0.1540], + [-0.1011, 0.0600, 1.1090, -0.3545, 0.1284, 1.1484, -1.0120, + -1.3508, -0.9513], + [1.8948, 0.8627, -2.1359, 1.3740, -0.7499, 1.5019, 0.6919, + -0.0842, -0.4294]], + + [[-0.2802, 0.6941, -0.4788, -0.3845, 1.7752, 1.2950, -1.9490, + -1.4138, -0.8853], + [-1.3752, -0.5457, -0.5305, 0.4018, 0.2934, 0.7931, 2.3845, + -1.0726, 0.0364], + [0.3621, 0.2609, 0.1269, -0.5950, 0.7212, 0.5959, 1.6264, + -0.8836, -0.9320], + [0.2003, -1.0758, -1.1560, -0.6472, -1.7549, 0.1264, 0.6044, + -1.6857, 1.1571], + [1.4277, -0.4915, 0.4496, 2.2027, 0.0730, -3.1792, -0.5125, + 3.5837, 1.0184], + [1.6495, 1.7145, -0.2143, -0.1230, -0.2205, 0.8250, 0.4943, + -0.9025, 0.0864]]]) + qa_prediction = qa_prediction.permute(1, 2, 0) + pred1, pred2 = qa_prediction.split(1, dim=-1) + pred1 = pred1.squeeze(-1) + pred2 = pred2.squeeze(-1) + target1 = torch.LongTensor([3, 0, 2, 4, 4, 0]) + target2 = torch.LongTensor([4, 1, 6, 8, 7, 1]) + metric = ExtractiveQAMetric() + metric.evaluate(pred1, pred2, target1, target2) + result = metric.get_metric() + truth = {'EM': 62.5, 'f_1': 72.5, 'noAns-f_1': 50.0, 'noAns-EM': 50.0, 'hasAns-f_1': 95.0, 'hasAns-EM': 75.0} + for k, v in truth.items(): + self.assertTrue(k in result) + self.assertEqual(v, result[k]) diff --git a/test/core/test_utils.py b/test/core/test_utils.py index 363d5fa1..29645fb1 100644 --- a/test/core/test_utils.py +++ b/test/core/test_utils.py @@ -119,7 +119,8 @@ class TestCache(unittest.TestCase): def test_cache_save(self): try: start_time = time.time() - embed, vocab, d = process_data_1('test/data_for_tests/word2vec_test.txt', 'test/data_for_tests/cws_train') + embed, vocab, d = process_data_1('test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt', + 'test/data_for_tests/cws_train') end_time = time.time() pre_time = end_time - start_time with open('test/demo1.pkl', 'rb') as f: @@ -128,7 +129,8 @@ class TestCache(unittest.TestCase): for i in range(embed.shape[0]): self.assertListEqual(embed[i].tolist(), _embed[i].tolist()) start_time = time.time() - embed, vocab, d = process_data_1('test/data_for_tests/word2vec_test.txt', 'test/data_for_tests/cws_train') + embed, vocab, d = process_data_1('test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt', + 'test/data_for_tests/cws_train') end_time = time.time() read_time = end_time - start_time print("Read using {:.3f}, while prepare using:{:.3f}".format(read_time, pre_time)) @@ -139,7 +141,7 @@ class TestCache(unittest.TestCase): def test_cache_save_overwrite_path(self): try: start_time = time.time() - embed, vocab, d = process_data_1('test/data_for_tests/word2vec_test.txt', 'test/data_for_tests/cws_train', + embed, vocab, d = process_data_1('test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt', 'test/data_for_tests/cws_train', _cache_fp='test/demo_overwrite.pkl') end_time = time.time() pre_time = end_time - start_time @@ -149,7 +151,8 @@ class TestCache(unittest.TestCase): for i in range(embed.shape[0]): self.assertListEqual(embed[i].tolist(), _embed[i].tolist()) start_time = time.time() - embed, vocab, d = process_data_1('test/data_for_tests/word2vec_test.txt', 'test/data_for_tests/cws_train', + embed, vocab, d = process_data_1('test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt', + 'test/data_for_tests/cws_train', _cache_fp='test/demo_overwrite.pkl') end_time = time.time() read_time = end_time - start_time @@ -161,7 +164,8 @@ class TestCache(unittest.TestCase): def test_cache_refresh(self): try: start_time = time.time() - embed, vocab, d = process_data_1('test/data_for_tests/word2vec_test.txt', 'test/data_for_tests/cws_train', + embed, vocab, d = process_data_1('test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt', + 'test/data_for_tests/cws_train', _refresh=True) end_time = time.time() pre_time = end_time - start_time @@ -171,7 +175,8 @@ class TestCache(unittest.TestCase): for i in range(embed.shape[0]): self.assertListEqual(embed[i].tolist(), _embed[i].tolist()) start_time = time.time() - embed, vocab, d = process_data_1('test/data_for_tests/word2vec_test.txt', 'test/data_for_tests/cws_train', + embed, vocab, d = process_data_1('test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt', + 'test/data_for_tests/cws_train', _refresh=True) end_time = time.time() read_time = end_time - start_time diff --git a/test/data_for_tests/embedding/small_bert/config.json b/test/data_for_tests/embedding/small_bert/config.json new file mode 100644 index 00000000..3e516872 --- /dev/null +++ b/test/data_for_tests/embedding/small_bert/config.json @@ -0,0 +1,13 @@ +{ + "attention_probs_dropout_prob": 0.1, + "hidden_act": "gelu", + "hidden_dropout_prob": 0.1, + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 64, + "max_position_embeddings": 32, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "type_vocab_size": 2, + "vocab_size": 20 +} \ No newline at end of file diff --git a/test/data_for_tests/embedding/small_bert/small_pytorch_model.bin b/test/data_for_tests/embedding/small_bert/small_pytorch_model.bin new file mode 100644 index 00000000..fe968fb5 Binary files /dev/null and b/test/data_for_tests/embedding/small_bert/small_pytorch_model.bin differ diff --git a/test/data_for_tests/embedding/small_bert/vocab.txt b/test/data_for_tests/embedding/small_bert/vocab.txt new file mode 100644 index 00000000..565e67af --- /dev/null +++ b/test/data_for_tests/embedding/small_bert/vocab.txt @@ -0,0 +1,20 @@ +[PAD] +[UNK] +[CLS] +[SEP] +this +is +a +small +bert +model +vocab +file +and +only +twenty +line +for +the +whole +text diff --git a/test/data_for_tests/embedding/small_elmo/char.dic b/test/data_for_tests/embedding/small_elmo/char.dic new file mode 100644 index 00000000..74285f34 --- /dev/null +++ b/test/data_for_tests/embedding/small_elmo/char.dic @@ -0,0 +1,229 @@ +! 33 +" 34 +# 35 +$ 36 +% 37 +& 38 +' 39 +( 40 +) 41 +* 42 ++ 43 +, 44 +- 45 +. 46 +/ 47 +0 48 +1 49 +2 50 +3 51 +4 52 +5 53 +6 54 +7 55 +8 56 +9 57 +: 58 +; 59 +< 60 += 61 +> 62 +? 63 +@ 64 +A 65 +B 66 +C 67 +D 68 +E 69 +F 70 +G 71 +H 72 +I 73 +J 74 +K 75 +L 76 +M 77 +N 78 +O 79 +P 80 +Q 81 +R 82 +S 83 +T 84 +U 85 +V 86 +W 87 +X 88 +Y 89 +Z 90 +[ 91 +\ 92 +] 93 +^ 94 +_ 95 +` 96 +a 97 +b 98 +c 99 +d 100 +e 101 +f 102 +g 103 +h 104 +i 105 +j 106 +k 107 +l 108 +m 109 +n 110 +o 111 +p 112 +q 113 +r 114 +s 115 +t 116 +u 117 +v 118 +w 119 +x 120 +y 121 +z 122 +{ 123 +| 124 +} 125 +~ 126 + 127 +€ 128 + 129 +‚ 130 +ƒ 131 +„ 132 +† 134 +‡ 135 +ˆ 136 +‰ 137 +Š 138 +‹ 139 +Œ 140 + 141 +Ž 142 + 143 + 144 +‘ 145 +’ 146 +“ 147 +” 148 +• 149 +– 150 +— 151 +˜ 152 +™ 153 +š 154 +› 155 +œ 156 + 157 +ž 158 +Ÿ 159 +  160 +¡ 161 +¢ 162 +£ 163 +¤ 164 +¥ 165 +¦ 166 +§ 167 +¨ 168 +© 169 +ª 170 +« 171 +¬ 172 +­ 173 +® 174 +¯ 175 +° 176 +± 177 +² 178 +³ 179 +´ 180 +µ 181 +¶ 182 +· 183 +¸ 184 +¹ 185 +º 186 +» 187 +¼ 188 +½ 189 +¾ 190 +¿ 191 +À 192 +Á 193 + 194 +à 195 +Ä 196 +Å 197 +Æ 198 +Ç 199 +È 200 +É 201 +Ê 202 +Ë 203 +Ì 204 +Í 205 +Î 206 +Ï 207 +Ð 208 +Ñ 209 +Ò 210 +Ó 211 +Ô 212 +Õ 213 +Ö 214 +× 215 +Ø 216 +Ù 217 +Ú 218 +Û 219 +Ü 220 +Ý 221 +Þ 222 +ß 223 +à 224 +á 225 +â 226 +ã 227 +ä 228 +å 229 +æ 230 +ç 231 +è 232 +é 233 +ê 234 +ë 235 +ì 236 +í 237 +î 238 +ï 239 +ð 240 +ñ 241 +ò 242 +ó 243 +ô 244 +õ 245 +ö 246 +÷ 247 +ø 248 +ù 249 +ú 250 +û 251 +ü 252 +ý 253 +þ 254 +ÿ 255 + 256 + 257 + 258 + 259 + 260 + 1 + -1 diff --git a/test/data_for_tests/embedding/small_elmo/elmo_1x16_16_32cnn_1xhighway_options.json b/test/data_for_tests/embedding/small_elmo/elmo_1x16_16_32cnn_1xhighway_options.json new file mode 100644 index 00000000..9c02ef72 --- /dev/null +++ b/test/data_for_tests/embedding/small_elmo/elmo_1x16_16_32cnn_1xhighway_options.json @@ -0,0 +1,29 @@ +{ + "lstm": { + "use_skip_connections": true, + "projection_dim": 16, + "cell_clip": 3, + "proj_clip": 3, + "dim": 16, + "n_layers": 1 + }, + "char_cnn": { + "activation": "relu", + "filters": [ + [ + 1, + 16 + ], + [ + 2, + 16 + ] + ], + "n_highway": 1, + "embedding": { + "dim": 4 + }, + "n_characters": 262, + "max_characters_per_token": 50 + } +} diff --git a/test/data_for_tests/embedding/small_elmo/elmo_mini_for_testing.pkl b/test/data_for_tests/embedding/small_elmo/elmo_mini_for_testing.pkl new file mode 100644 index 00000000..4c72f3d5 Binary files /dev/null and b/test/data_for_tests/embedding/small_elmo/elmo_mini_for_testing.pkl differ diff --git a/test/data_for_tests/glove.6B.50d_test.txt b/test/data_for_tests/embedding/small_static_embedding/glove.6B.50d_test.txt similarity index 100% rename from test/data_for_tests/glove.6B.50d_test.txt rename to test/data_for_tests/embedding/small_static_embedding/glove.6B.50d_test.txt diff --git a/test/data_for_tests/word2vec_test.txt b/test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt similarity index 100% rename from test/data_for_tests/word2vec_test.txt rename to test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt diff --git a/test/data_for_tests/io/cws_msra/dev.txt b/test/data_for_tests/io/cws_msra/dev.txt new file mode 100644 index 00000000..9c6b34ee --- /dev/null +++ b/test/data_for_tests/io/cws_msra/dev.txt @@ -0,0 +1,2 @@ +“ 人们 常 说 生活 是 一 部 教科书 ， 而 血 与 火 的 战争 更 是 不可多得 的 教科书 ， 她 确实 是 名副其实 的 ‘ 我 的 大学 ’ 。 +他 “ 严格要求 自己 ， 从 一个 科举 出身 的 进士 成为 一个 伟大 的 民主主义 者 ， 进而 成为 一 位 杰出 的 党外 共产主义 战士 ， 献身 于 崇高 的 共产主义 事业 。 diff --git a/test/data_for_tests/io/cws_msra/test.txt b/test/data_for_tests/io/cws_msra/test.txt new file mode 100644 index 00000000..8d5c6b3c --- /dev/null +++ b/test/data_for_tests/io/cws_msra/test.txt @@ -0,0 +1,2 @@ +扬帆 远东 做 与 中国 合作 的 先行 +希腊 的 经济 结构 较 特殊 。 diff --git a/test/data_for_tests/io/cws_msra/train.txt b/test/data_for_tests/io/cws_msra/train.txt new file mode 100644 index 00000000..35c2cad0 --- /dev/null +++ b/test/data_for_tests/io/cws_msra/train.txt @@ -0,0 +1,3 @@ +“ 心 静 渐 知 春 似 海 ， 花 深 每 觉 影 生 香 。 +“ 吃 屎 的 东西 ， 连 一 捆 麦 也 铡 不 动 呀 ？ +复旦大学 百年 校庆 。 \ No newline at end of file diff --git a/test/data_for_tests/io/imdb/dev.txt b/test/data_for_tests/io/imdb/dev.txt new file mode 100644 index 00000000..6b548a0c --- /dev/null +++ b/test/data_for_tests/io/imdb/dev.txt @@ -0,0 +1,2 @@ +neg It, at all, you have seen when harry met sally, then avoid this one. It will not only make you bang your head on the table as why can't bollywood even make a good remake; but also annoy you with the so called funny moments in it. The charm of the movie is missing. Ranee looks terrible. Saif tries to act like he is one hell of an actor. The plots that have been picked up from the original, don't look effective either. The part where both of them bring their friends along and they hit a note, it just doesn't look appealing. What can be more disastrous? you wanna waste some money, this is what you can get. Otherwise, put some more bucks, and watch the original. Its too good to miss.. +neg The monster from Enemy Mine somehow made his way into a small mountain community, where he has taken up residence. He's being hunted by a female doctor-turned-vigilante who is out to exterminate him. This female assassin, who looks like a refugee from a Motley Crue video, rides around on a motorcycle and tries to save a bunch of kids who have chosen to have a Big Chill weekend right smack dab in the middle of the monster's turf. Decapitations and lots of blood are primarily in place to draw attention away from the story which limps along like a bad version of the Island of Dr. Moreau (and yes, it's worse than the one with Val Kilmer). diff --git a/test/data_for_tests/io/imdb/test.txt b/test/data_for_tests/io/imdb/test.txt new file mode 100644 index 00000000..c9bfae74 --- /dev/null +++ b/test/data_for_tests/io/imdb/test.txt @@ -0,0 +1,2 @@ +neg Alan Rickman & Emma Thompson give good performances with southern/New Orleans accents in this detective flick. It's worth seeing for their scenes- and Rickman's scene with Hal Holbrook. These three actors mannage to entertain us no matter what the movie, it seems. The plot for the movie shows potential, but one gets the impression in watching the film that it was not pulled off as well as it could have been. The fact that it is cluttered by a rather uninteresting subplot and mostly uninteresting kidnappers really muddles things. The movie is worth a view- if for nothing more than entertaining performances by Rickman, Thompson, and Holbrook. +neg I have seen this movie and I did not care for this movie anyhow. I would not think about going to Paris because I do not like this country and its national capital. I do not like to learn french anyhow because I do not understand their language. Why would I go to France when I rather go to Germany or the United Kingdom? Germany and the United Kingdom are the nations I tolerate. Apparently the Olsen Twins do not understand the French language just like me. Therefore I will not bother the France trip no matter what. I might as well stick to the United Kingdom and meet single women and play video games if there is a video arcade. That is all. diff --git a/test/data_for_tests/io/imdb/train.txt b/test/data_for_tests/io/imdb/train.txt new file mode 100644 index 00000000..d6ac6b68 --- /dev/null +++ b/test/data_for_tests/io/imdb/train.txt @@ -0,0 +1,2 @@ +neg I'll try to use words to describe this on....

I saw the original, which was good in its own way, but back then I should have feared a sequel.

And I was 'afraid' when I picked this one up, but now that I've seen it, I have to say, it's even worse then I thought. Why these movies still get money still makes my mind spin.

Let's start with the actors;they aren't all that good, but it has to be said, some make heads turn by being just plain awful. But what can an actor do with a script like this one. It's trying to be a copy of the original only this time the places have changed, any form of story is gone and any attempt of actually coming up with something that hasn't been done before, fails miserably. In a futile attempt to get it up-to-date, they try to make it exciting by making use of the whole 'big-brother' theme , but that has been worn out ages ago and offers nothing but a filler for between the beginning and the end. An attempt was made to try to save the movie by making a ton of references to the '83 original, but it just ended up being plain funny and sometimes a bit sad. In conclusion, if you have nothing , and I mean nothing , to do... go watch it, or play Frisbee... with the DVD.... by yourself. It'll offer you the same amount of fun.. I promise +pos This movie is totally wicked! It's really great to see MJH in a different role than her Sabrina character! The plot is totally cool, and the characters are excellently written. Definitely one of the best movies!! diff --git a/test/data_for_tests/io/rte/dev.tsv b/test/data_for_tests/io/rte/dev.tsv new file mode 100644 index 00000000..725d7542 --- /dev/null +++ b/test/data_for_tests/io/rte/dev.tsv @@ -0,0 +1,3 @@ +index sentence1 sentence2 label +0 Dana Reeve, the widow of the actor Christopher Reeve, has died of lung cancer at age 44, according to the Christopher Reeve Foundation. Christopher Reeve had an accident. not_entailment +1 Yet, we now are discovering that antibiotics are losing their effectiveness against illness. Disease-causing bacteria are mutating faster than we can come up with new antibiotics to fight the new variations. Bacteria is winning the war against antibiotics. entailment diff --git a/test/data_for_tests/io/rte/test.tsv b/test/data_for_tests/io/rte/test.tsv new file mode 100644 index 00000000..aeceb467 --- /dev/null +++ b/test/data_for_tests/io/rte/test.tsv @@ -0,0 +1,3 @@ +index sentence1 sentence2 +0 Mangla was summoned after Madhumita's sister Nidhi Shukla, who was the first witness in the case. Shukla is related to Mangla. +1 Authorities in Brazil say that more than 200 people are being held hostage in a prison in the country's remote, Amazonian-jungle state of Rondonia. Authorities in Brazil hold 200 people as hostage. diff --git a/test/data_for_tests/io/rte/train.tsv b/test/data_for_tests/io/rte/train.tsv new file mode 100644 index 00000000..9f3dab6e --- /dev/null +++ b/test/data_for_tests/io/rte/train.tsv @@ -0,0 +1,4 @@ +index sentence1 sentence2 label +0 No Weapons of Mass Destruction Found in Iraq Yet. Weapons of Mass Destruction Found in Iraq. not_entailment +1 A place of sorrow, after Pope John Paul II died, became a place of celebration, as Roman Catholic faithful gathered in downtown Chicago to mark the installation of new Pope Benedict XVI. Pope Benedict XVI is the new leader of the Roman Catholic Church. entailment +2 Herceptin was already approved to treat the sickest breast cancer patients, and the company said, Monday, it will discuss with federal regulators the possibility of prescribing the drug for more breast cancer patients. Herceptin can be used to treat breast cancer. entailment diff --git a/test/data_for_tests/sample_mnli.tsv b/test/data_for_tests/sample_mnli.tsv new file mode 100644 index 00000000..9a30b95b --- /dev/null +++ b/test/data_for_tests/sample_mnli.tsv @@ -0,0 +1,12 @@ +index promptID pairID genre sentence1_binary_parse sentence2_binary_parse sentence1_parse sentence2_parse sentence1 sentence2 label1 label2 label3 label4 label5 gold_label +0 63735 63735n slate ( ( The ( new rights ) ) ( are ( nice enough ) ) ) ( Everyone ( really ( likes ( the ( newest benefits ) ) ) ) ) (ROOT (S (NP (DT The) (JJ new) (NNS rights)) (VP (VBP are) (ADJP (JJ nice) (RB enough))))) (ROOT (S (NP (NN Everyone)) (VP (ADVP (RB really)) (VBZ likes) (NP (DT the) (JJS newest) (NNS benefits))))) The new rights are nice enough Everyone really likes the newest benefits neutral entailment neutral neutral neutral neutral +1 91383 91383c government ( ( This site ) ( ( includes ( ( ( ( a list ) ( of ( all ( award winners ) ) ) ) and ) ( ( a ( searchable database ) ) ( of ( Government ( Executive articles ) ) ) ) ) ) . ) ) ( ( ( The ( Government ( Executive articles ) ) ) ( housed ( on ( the website ) ) ) ) ( ( ( are not ) ( able ( to ( be searched ) ) ) ) . ) ) (ROOT (S (NP (DT This) (NN site)) (VP (VBZ includes) (NP (NP (NP (DT a) (NN list)) (PP (IN of) (NP (DT all) (NN award) (NNS winners)))) (CC and) (NP (NP (DT a) (JJ searchable) (NN database)) (PP (IN of) (NP (NNP Government) (NNP Executive) (NNS articles)))))) (. .))) (ROOT (S (NP (NP (DT The) (NNP Government) (NNP Executive) (NNS articles)) (VP (VBN housed) (PP (IN on) (NP (DT the) (NN website))))) (VP (VBP are) (RB not) (ADJP (JJ able) (S (VP (TO to) (VP (VB be) (ADJP (JJ searched))))))) (. .))) This site includes a list of all award winners and a searchable database of Government Executive articles. The Government Executive articles housed on the website are not able to be searched. contradiction contradiction contradiction contradiction contradiction contradiction +2 755 755e telephone ( ( ( ( uh ( i ( ( do n't ) ( know ( ( i i ) ( have ( ( mixed emotions ) ( about ( him ( ( uh sometimes ) ( i ( like him ) ) ) ) ) ) ) ) ) ) ) ) but ) ( ( at ( the ( same times ) ) ) ( i ( love ( to ( see somebody ) ) ) ) ) ) ( beat him ) ) ( I ( ( ( ( ( ( like him ) ( for ( the ( most part ) ) ) ) , ) but ) ( ( would still ) ( enjoy ( seeing ( someone ( beat him ) ) ) ) ) ) . ) ) (ROOT (SINV (S (S (INTJ (UH uh)) (NP (FW i)) (VP (VBP do) (RB n't) (VP (VB know) (NP (NP (FW i) (FW i)) (SBAR (S (VP (VBP have) (VP (VBN mixed) (NP (NNS emotions)) (PP (IN about) (S (NP (PRP him)) (VP (VBG uh) (ADVP (RB sometimes)) (NP (NP (FW i)) (PP (IN like) (NP (PRP him))))))))))))))) (CC but) (S (PP (IN at) (NP (DT the) (JJ same) (NNS times))) (NP (FW i)) (VP (VBP love) (S (VP (TO to) (VP (VB see) (NP (NN somebody)))))))) (VP (VBD beat)) (NP (PRP him)))) (ROOT (S (NP (PRP I)) (VP (VP (VBP like) (NP (PRP him)) (PP (IN for) (NP (DT the) (JJS most) (NN part)))) (, ,) (CC but) (VP (MD would) (ADVP (RB still)) (VP (VB enjoy) (S (VP (VBG seeing) (S (NP (NN someone)) (VP (VB beat) (NP (PRP him))))))))) (. .))) uh i don't know i i have mixed emotions about him uh sometimes i like him but at the same times i love to see somebody beat him I like him for the most part, but would still enjoy seeing someone beat him. entailment entailment entailment entailment entailment entailment +3 78013 78013c telephone ( yeah ( ( i i ) ( think ( ( my ( favorite restaurant ) ) ( ( is always ) ( been ( ( the ( one closest ) ) ( you ( ( know ( the closest ) ) ( ( as long ) ( as ( it ( 's ( it ( meets ( ( the ( minimum criteria ) ) ( you ( know ( of ( good food ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ( ( My ( favorite restaurants ) ) ( ( ( ( are always ) ( ( ( ( ( at least ) a ) hundred ) miles ) away ) ) ( from ( my house ) ) ) . ) ) (ROOT (S (VP (VB yeah) (NP (NP (FW i) (FW i)) (SBAR (S (VP (VBP think) (SBAR (S (NP (PRP$ my) (JJ favorite) (NN restaurant)) (VP (VBZ is) (ADVP (RB always)) (VP (VBN been) (NP (NP (DT the) (CD one) (JJS closest)) (SBAR (S (NP (PRP you)) (VP (VBP know) (NP (DT the) (JJS closest)) (ADVP (ADVP (RB as) (RB long)) (SBAR (IN as) (S (NP (PRP it)) (VP (VBZ 's) (SBAR (S (NP (PRP it)) (VP (VBZ meets) (NP (NP (DT the) (JJ minimum) (NNS criteria)) (SBAR (S (NP (PRP you)) (VP (VBP know) (PP (IN of) (NP (JJ good) (NN food))))))))))))))))))))))))))))) (ROOT (S (NP (PRP$ My) (JJ favorite) (NNS restaurants)) (VP (VBP are) (ADVP (RB always)) (ADVP (NP (QP (IN at) (JJS least) (DT a) (CD hundred)) (NNS miles)) (RB away)) (PP (IN from) (NP (PRP$ my) (NN house)))) (. .))) yeah i i think my favorite restaurant is always been the one closest you know the closest as long as it's it meets the minimum criteria you know of good food My favorite restaurants are always at least a hundred miles away from my house. contradiction contradiction contradiction contradiction contradiction contradiction +4 96377 96377c telephone ( i ( ( do n't ) ( know ( um ( do ( you ( do ( ( a lot ) ( of camping ) ) ) ) ) ) ) ) ) ( I ( ( know exactly ) . ) ) (ROOT (S (NP (FW i)) (VP (VBP do) (RB n't) (VP (VB know) (SBAR (S (NP (NN um)) (VP (VBP do) (SBAR (S (NP (PRP you)) (VP (VBP do) (NP (NP (DT a) (NN lot)) (PP (IN of) (NP (NN camping)))))))))))))) (ROOT (S (NP (PRP I)) (VP (VBP know) (ADVP (RB exactly))) (. .))) i don't know um do you do a lot of camping I know exactly. contradiction contradiction contradiction contradiction contradiction contradiction +5 139749 139749c telephone ( well ( that ( would ( be ( ( a help ) ( i ( wish ( they ( would ( do ( that ( ( ( here ( we ( have ( got ( so ( ( little ( landfill space ) ) ( left ( that ( we ( 're ( going ( to ( ( run out ) ( before ( ( the end ) ( of ( this decade ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) and ) ( it ( ( 's really ) ( going ( to be ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ( We ( ( have ( plenty ( of ( space ( in ( the landfill ) ) ) ) ) ) . ) ) (ROOT (FRAG (ADVP (RB well)) (SBAR (WHNP (WDT that)) (S (VP (MD would) (VP (VB be) (NP (NP (DT a) (NN help)) (SBAR (S (NP (FW i)) (VP (VBP wish) (SBAR (S (NP (PRP they)) (VP (MD would) (VP (VB do) (SBAR (IN that) (S (S (ADVP (RB here)) (NP (PRP we)) (VP (VBP have) (VP (VBN got) (SBAR (IN so) (S (NP (JJ little) (NN landfill) (NN space)) (VP (VBD left) (SBAR (IN that) (S (NP (PRP we)) (VP (VBP 're) (VP (VBG going) (S (VP (TO to) (VP (VB run) (PRT (RP out)) (PP (IN before) (NP (NP (DT the) (NN end)) (PP (IN of) (NP (DT this) (NN decade)))))))))))))))))) (CC and) (S (NP (PRP it)) (VP (VBZ 's) (ADVP (RB really)) (VP (VBG going) (S (VP (TO to) (VP (VB be))))))))))))))))))))))) (ROOT (S (NP (PRP We)) (VP (VBP have) (NP (NP (RB plenty)) (PP (IN of) (NP (NP (NN space)) (PP (IN in) (NP (DT the) (NN landfill))))))) (. .))) well that would be a help i wish they would do that here we have got so little landfill space left that we're going to run out before the end of this decade and it's really going to be We have plenty of space in the landfill. contradiction contradiction contradiction contradiction contradiction contradiction +6 101415 101415c telephone ( yeah ( ( ( i know ) and ) ( i ( did ( that ( ( ( all ( through college ) ) and ) ( it ( worked too ) ) ) ) ) ) ) ) ( I ( ( ( did ( that all ) ) ( through college ) ) ( but ( it ( never worked ) ) ) ) ) (ROOT (S (VP (VB yeah) (S (S (NP (FW i)) (VP (VBP know))) (CC and) (S (NP (FW i)) (VP (VBD did) (SBAR (IN that) (S (S (NP (DT all)) (PP (IN through) (NP (NN college)))) (CC and) (S (NP (PRP it)) (VP (VBD worked) (ADVP (RB too)))))))))))) (ROOT (S (NP (PRP I)) (VP (VBD did) (ADVP (IN that) (DT all)) (PP (IN through) (NP (NN college))) (SBAR (CC but) (S (NP (PRP it)) (ADVP (RB never)) (VP (VBD worked))))))) yeah i know and i did that all through college and it worked too I did that all through college but it never worked contradiction contradiction contradiction contradiction contradiction contradiction +7 93958 93958n travel ( ( ( ( ( Calcutta ( seems ( to ( be ( ( the ( only ( other ( production center ) ) ) ) ( ( having ( any pretensions ) ) ( to ( ( artistic creativity ) ( at all ) ) ) ) ) ) ) ) ) , ) but ) ( ironically ( you ( ( 're actually ) ( ( more ( likely ( to ( see ( ( the works ) ( of ( ( ( Satyajit Ray ) or ) ( ( Mrinal Sen ) ( shown ( in ( Europe ( or ( North America ) ) ) ) ) ) ) ) ) ) ) ) ) ( than ( in ( India itself ) ) ) ) ) ) ) ) . ) ( ( Most ( of ( ( Mrinal ( Sen 's ) ) work ) ) ) ( ( can ( be ( found ( in ( European collections ) ) ) ) ) . ) ) (ROOT (S (S (NP (NNP Calcutta)) (VP (VBZ seems) (S (VP (TO to) (VP (VB be) (NP (NP (DT the) (JJ only) (JJ other) (NN production) (NN center)) (VP (VBG having) (NP (DT any) (NNS pretensions)) (PP (TO to) (NP (NP (JJ artistic) (NN creativity)) (ADVP (IN at) (DT all))))))))))) (, ,) (CC but) (S (ADVP (RB ironically)) (NP (PRP you)) (VP (VBP 're) (ADVP (RB actually)) (ADJP (ADJP (RBR more) (JJ likely) (S (VP (TO to) (VP (VB see) (NP (NP (DT the) (NNS works)) (PP (IN of) (NP (NP (NNP Satyajit) (NNP Ray)) (CC or) (NP (NP (NNP Mrinal) (NNP Sen)) (VP (VBN shown) (PP (IN in) (NP (NNP Europe) (CC or) (NNP North) (NNP America)))))))))))) (ADVP (IN than) (PP (IN in) (S (VP (VBG India) (NP (PRP itself))))))))) (. .))) (ROOT (S (NP (NP (JJS Most)) (PP (IN of) (NP (NP (NNP Mrinal) (NNP Sen) (POS 's)) (NN work)))) (VP (MD can) (VP (VB be) (VP (VBN found) (PP (IN in) (NP (JJ European) (NNS collections)))))) (. .))) Calcutta seems to be the only other production center having any pretensions to artistic creativity at all, but ironically you're actually more likely to see the works of Satyajit Ray or Mrinal Sen shown in Europe or North America than in India itself. Most of Mrinal Sen's work can be found in European collections. neutral neutral entailment neutral neutral neutral +8 12567 12567c slate ( ( If ( ( that investor ) ( were ( willing ( to ( pay ( extra ( for ( ( the security ) ( of ( limited downside ) ) ) ) ) ) ) ) ) ) ) ( , ( she ( ( could ( ( buy ( put options ) ) ( with ( ( a ( strike price ) ) ( of ( ( ( $ 98 ) , ) ( which ( would ( ( ( lock ( in ( ( her profit ) ( on ( ( the shares ) ( at ( $ 18 ) ) ) ) ) ) ) , ) ( less ( whatever ( ( the options ) cost ) ) ) ) ) ) ) ) ) ) ) ) . ) ) ) ) ( ( THe ( strike price ) ) ( ( could ( be ( $ 8 ) ) ) . ) ) (ROOT (S (SBAR (IN If) (S (NP (DT that) (NN investor)) (VP (VBD were) (ADJP (JJ willing) (S (VP (TO to) (VP (VB pay) (NP (NP (JJ extra)) (PP (IN for) (NP (NP (DT the) (NN security)) (PP (IN of) (NP (JJ limited) (NN downside))))))))))))) (, ,) (NP (PRP she)) (VP (MD could) (VP (VB buy) (NP (NN put) (NNS options)) (PP (IN with) (NP (NP (DT a) (NN strike) (NN price)) (PP (IN of) (NP (NP ($ $) (CD 98)) (, ,) (SBAR (WHNP (WDT which)) (S (VP (MD would) (VP (VB lock) (PP (IN in) (NP (NP (PRP$ her) (NN profit)) (PP (IN on) (NP (NP (DT the) (NNS shares)) (PP (IN at) (NP ($ $) (CD 18))))))) (, ,) (ADVP (ADVP (RBR less)) (SBAR (WHNP (WDT whatever)) (S (NP (DT the) (NNS options)) (VP (VBD cost))))))))))))))) (. .))) (ROOT (S (NP (NNP THe) (NN strike) (NN price)) (VP (MD could) (VP (VB be) (NP ($ $) (CD 8)))) (. .))) If that investor were willing to pay extra for the security of limited downside, she could buy put options with a strike price of $98, which would lock in her profit on the shares at $18, less whatever the options cost. THe strike price could be $8. contradiction contradiction contradiction contradiction contradiction contradiction +9 117487 117487n slate ( ( 3 -RRB- ) ( ( Dare ( you ( ( ( rise ( to ( ( ( ( the occasion ) , ) ( like Raskolnikov ) ) , ) ) ) and ) ( reject ( ( the ( petty rules ) ) ( that ( govern ( lesser men ) ) ) ) ) ) ) ) ? ) ) ( ( ( Would you ) ( ( ( rise up ) and ) ( defeaat ( ( all ( evil lords ) ) ( in ( the town ) ) ) ) ) ) ? ) (ROOT (S (LST (LS 3) (-RRB- -RRB-)) (VP (VB Dare) (S (NP (PRP you)) (VP (VP (VB rise) (PP (TO to) (NP (NP (DT the) (NN occasion)) (, ,) (PP (IN like) (NP (NNP Raskolnikov))) (, ,)))) (CC and) (VP (VB reject) (NP (NP (DT the) (JJ petty) (NNS rules)) (SBAR (WHNP (WDT that)) (S (VP (VBP govern) (NP (JJR lesser) (NNS men)))))))))) (. ?))) (ROOT (SQ (MD Would) (NP (PRP you)) (VP (VP (VB rise) (PRT (RP up))) (CC and) (VP (VB defeaat) (NP (NP (DT all) (JJ evil) (NNS lords)) (PP (IN in) (NP (DT the) (NN town)))))) (. ?))) 3) Dare you rise to the occasion, like Raskolnikov, and reject the petty rules that govern lesser men? Would you rise up and defeaat all evil lords in the town? neutral neutral neutral neutral neutral neutral +10 9616 9616c travel ( ( The ( ( most important ) directions ) ) ( ( ( are ( simply ( ( up and ) up ) ) ) ( ( ( ( ( ( ( ( leads eventually ) ( to ( the cathedral ) ) ) and ) ( fortress ( commanding ( the hilltop ) ) ) ) , ) and ) down ) ( inevitably ( ( leads ( to ( one ( of ( three gates ) ) ) ) ) ( through ( ( the wall ) ( to ( the ( new town ) ) ) ) ) ) ) ) ) . ) ) ( Go ( ( downwards ( to ( one ( of ( ( ( the gates ) , ) ( ( all ( of which ) ) ( will ( ( lead you ) ( into ( the cathedral ) ) ) ) ) ) ) ) ) ) . ) ) (ROOT (S (NP (DT The) (ADJP (RBS most) (JJ important)) (NNS directions)) (VP (VBP are) (PRN (ADVP (RB simply)) (ADVP (RB up) (CC and) (RB up))) (VP (VP (VBZ leads) (ADVP (RB eventually)) (PP (TO to) (NP (DT the) (NN cathedral)))) (CC and) (VP (VBZ fortress) (NP (JJ commanding) (DT the) (NN hilltop))) (, ,) (CC and) (ADVP (RB down)) (VP (ADVP (RB inevitably)) (VBZ leads) (PP (TO to) (NP (NP (CD one)) (PP (IN of) (NP (CD three) (NNS gates))))) (PP (IN through) (NP (NP (DT the) (NN wall)) (PP (TO to) (NP (DT the) (JJ new) (NN town)))))))) (. .))) (ROOT (S (NP (NNP Go)) (VP (VBZ downwards) (PP (TO to) (NP (NP (CD one)) (PP (IN of) (NP (NP (DT the) (NNS gates)) (, ,) (SBAR (WHNP (DT all) (WHPP (IN of) (WHNP (WDT which)))) (S (VP (MD will) (VP (VB lead) (NP (PRP you)) (PP (IN into) (NP (DT the) (NN cathedral)))))))))))) (. .))) The most important directions are simply up and up leads eventually to the cathedral and fortress commanding the hilltop, and down inevitably leads to one of three gates through the wall to the new town. Go downwards to one of the gates, all of which will lead you into the cathedral. contradiction contradiction entailment contradiction contradiction contradiction diff --git a/test/embeddings/test_bert_embedding.py b/test/embeddings/test_bert_embedding.py new file mode 100644 index 00000000..71511458 --- /dev/null +++ b/test/embeddings/test_bert_embedding.py @@ -0,0 +1,39 @@ +import unittest +from fastNLP import Vocabulary +from fastNLP.embeddings import BertEmbedding +import torch +import os + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestDownload(unittest.TestCase): + def test_download(self): + # import os + vocab = Vocabulary().add_word_lst("This is a test .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='en') + words = torch.LongTensor([[2, 3, 4, 0]]) + print(embed(words).size()) + + for pool_method in ['first', 'last', 'max', 'avg']: + for include_cls_sep in [True, False]: + embed = BertEmbedding(vocab, model_dir_or_name='en', pool_method=pool_method, + include_cls_sep=include_cls_sep) + print(embed(words).size()) + + def test_word_drop(self): + vocab = Vocabulary().add_word_lst("This is a test .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='en', dropout=0.1, word_dropout=0.2) + for i in range(10): + words = torch.LongTensor([[2, 3, 4, 0]]) + print(embed(words).size()) + + +class TestBertEmbedding(unittest.TestCase): + def test_bert_embedding_1(self): + vocab = Vocabulary().add_word_lst("this is a test . [SEP]".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', word_dropout=0.1) + requires_grad = embed.requires_grad + embed.requires_grad = not requires_grad + embed.train() + words = torch.LongTensor([[2, 3, 4, 0]]) + result = embed(words) + self.assertEqual(result.size(), (1, 4, 16)) diff --git a/test/embeddings/test_elmo_embedding.py b/test/embeddings/test_elmo_embedding.py new file mode 100644 index 00000000..bfb31659 --- /dev/null +++ b/test/embeddings/test_elmo_embedding.py @@ -0,0 +1,36 @@ + +import unittest +from fastNLP import Vocabulary +from fastNLP.embeddings import ElmoEmbedding +import torch +import os + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestDownload(unittest.TestCase): + def test_download_small(self): + # import os + vocab = Vocabulary().add_word_lst("This is a test .".split()) + elmo_embed = ElmoEmbedding(vocab, model_dir_or_name='en-small') + words = torch.LongTensor([[0, 1, 2]]) + print(elmo_embed(words).size()) + + +# 首先保证所有权重可以加载；上传权重；验证可以下载 + + +class TestRunElmo(unittest.TestCase): + def test_elmo_embedding(self): + vocab = Vocabulary().add_word_lst("This is a test .".split()) + elmo_embed = ElmoEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_elmo', layers='0,1') + words = torch.LongTensor([[0, 1, 2]]) + hidden = elmo_embed(words) + print(hidden.size()) + + def test_elmo_embedding_layer_assertion(self): + vocab = Vocabulary().add_word_lst("This is a test .".split()) + try: + elmo_embed = ElmoEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_elmo', + layers='0,1,2') + except AssertionError as e: + print(e) + diff --git a/test/embeddings/test_static_embedding.py b/test/embeddings/test_static_embedding.py index 0c8fc739..7d1e8302 100644 --- a/test/embeddings/test_static_embedding.py +++ b/test/embeddings/test_static_embedding.py @@ -3,13 +3,140 @@ import unittest from fastNLP.embeddings import StaticEmbedding from fastNLP import Vocabulary import torch +import os + + +class TestLoad(unittest.TestCase): + def test_norm1(self): + # 测试只对可以找到的norm + vocab = Vocabulary().add_word_lst(['the', 'a', 'notinfile']) + embed = StaticEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_static_embedding/' + 'glove.6B.50d_test.txt', + only_norm_found_vector=True) + self.assertEqual(round(torch.norm(embed(torch.LongTensor([[2]]))).item(), 4), 1) + self.assertNotEqual(torch.norm(embed(torch.LongTensor([[4]]))).item(), 1) + + def test_norm2(self): + # 测试对所有都norm + vocab = Vocabulary().add_word_lst(['the', 'a', 'notinfile']) + embed = StaticEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_static_embedding/' + 'glove.6B.50d_test.txt', + normalize=True) + self.assertEqual(round(torch.norm(embed(torch.LongTensor([[2]]))).item(), 4), 1) + self.assertEqual(round(torch.norm(embed(torch.LongTensor([[4]]))).item(), 4), 1) + + def test_dropword(self): + # 测试是否可以通过drop word + vocab = Vocabulary().add_word_lst([chr(i) for i in range(1, 200)]) + embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=10, dropout=0.1, word_dropout=0.4) + for i in range(10): + length = torch.randint(1, 50, (1,)).item() + batch = torch.randint(1, 4, (1,)).item() + words = torch.randint(1, 200, (batch, length)).long() + embed(words) class TestRandomSameEntry(unittest.TestCase): def test_same_vector(self): - vocab = Vocabulary().add_word_lst(["The", "the", "THE"]) + vocab = Vocabulary().add_word_lst(["The", "the", "THE", 'a', "A"]) embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=5, lower=True) - words = torch.LongTensor([[vocab.to_index(word) for word in ["The", "the", "THE"]]]) + words = torch.LongTensor([[vocab.to_index(word) for word in ["The", "the", "THE", 'a', 'A']]]) words = embed(words) embed_0 = words[0, 0] - for i in range(1, words.size(1)): + for i in range(1, 3): assert torch.sum(embed_0==words[0, i]).eq(len(embed_0)) + embed_0 = words[0, 3] + for i in range(3, 5): + assert torch.sum(embed_0 == words[0, i]).eq(len(embed_0)) + + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_same_vector2(self): + vocab = Vocabulary().add_word_lst(["The", 'a', 'b', "the", "THE", "B", 'a', "A"]) + embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6B-100d', + lower=True) + words = torch.LongTensor([[vocab.to_index(word) for word in ["The", "the", "THE", 'b', "B", 'a', 'A']]]) + words = embed(words) + embed_0 = words[0, 0] + for i in range(1, 3): + assert torch.sum(embed_0==words[0, i]).eq(len(embed_0)) + embed_0 = words[0, 3] + for i in range(3, 5): + assert torch.sum(embed_0 == words[0, i]).eq(len(embed_0)) + + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_same_vector3(self): + # 验证lower + word_lst = ["The", "the"] + no_create_word_lst = ['of', 'Of', 'With', 'with'] + vocab = Vocabulary().add_word_lst(word_lst) + vocab.add_word_lst(no_create_word_lst, no_create_entry=True) + embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6B-100d', + lower=True) + words = torch.LongTensor([[vocab.to_index(word) for word in word_lst+no_create_word_lst]]) + words = embed(words) + + lowered_word_lst = [word.lower() for word in word_lst] + lowered_no_create_word_lst = [word.lower() for word in no_create_word_lst] + lowered_vocab = Vocabulary().add_word_lst(lowered_word_lst) + lowered_vocab.add_word_lst(lowered_no_create_word_lst, no_create_entry=True) + lowered_embed = StaticEmbedding(lowered_vocab, model_dir_or_name='en-glove-6B-100d', + lower=False) + lowered_words = torch.LongTensor([[lowered_vocab.to_index(word) for word in lowered_word_lst+lowered_no_create_word_lst]]) + lowered_words = lowered_embed(lowered_words) + + all_words = word_lst + no_create_word_lst + + for idx, (word_i, word_j) in enumerate(zip(words[0], lowered_words[0])): + with self.subTest(idx=idx, word=all_words[idx]): + assert torch.sum(word_i == word_j).eq(lowered_embed.embed_size) + + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_same_vector4(self): + # 验证在有min_freq下的lower + word_lst = ["The", "the", "the", "The", "a", "A"] + no_create_word_lst = ['of', 'Of', "Of", "of", 'With', 'with'] + all_words = word_lst[:-2] + no_create_word_lst[:-2] + vocab = Vocabulary(min_freq=2).add_word_lst(word_lst) + vocab.add_word_lst(no_create_word_lst, no_create_entry=True) + embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6B-100d', + lower=True) + words = torch.LongTensor([[vocab.to_index(word) for word in all_words]]) + words = embed(words) + + lowered_word_lst = [word.lower() for word in word_lst] + lowered_no_create_word_lst = [word.lower() for word in no_create_word_lst] + lowered_vocab = Vocabulary().add_word_lst(lowered_word_lst) + lowered_vocab.add_word_lst(lowered_no_create_word_lst, no_create_entry=True) + lowered_embed = StaticEmbedding(lowered_vocab, model_dir_or_name='en-glove-6B-100d', + lower=False) + lowered_words = torch.LongTensor([[lowered_vocab.to_index(word.lower()) for word in all_words]]) + lowered_words = lowered_embed(lowered_words) + + for idx in range(len(all_words)): + word_i, word_j = words[0, idx], lowered_words[0, idx] + with self.subTest(idx=idx, word=all_words[idx]): + assert torch.sum(word_i == word_j).eq(lowered_embed.embed_size) + + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_same_vector5(self): + # 检查通过使用min_freq后的word是否内容一致 + word_lst = ["they", "the", "they", "the", 'he', 'he', "a", "A"] + no_create_word_lst = ['of', "of", "she", "she", 'With', 'with'] + all_words = word_lst[:-2] + no_create_word_lst[:-2] + vocab = Vocabulary().add_word_lst(word_lst) + vocab.add_word_lst(no_create_word_lst, no_create_entry=True) + embed = StaticEmbedding(vocab, model_dir_or_name='en-glove-6B-100d', + lower=False, min_freq=2) + words = torch.LongTensor([[vocab.to_index(word) for word in all_words]]) + words = embed(words) + + min_freq_vocab = Vocabulary(min_freq=2).add_word_lst(word_lst) + min_freq_vocab.add_word_lst(no_create_word_lst, no_create_entry=True) + min_freq_embed = StaticEmbedding(min_freq_vocab, model_dir_or_name='en-glove-6B-100d', + lower=False) + min_freq_words = torch.LongTensor([[min_freq_vocab.to_index(word.lower()) for word in all_words]]) + min_freq_words = min_freq_embed(min_freq_words) + + for idx in range(len(all_words)): + word_i, word_j = words[0, idx], min_freq_words[0, idx] + with self.subTest(idx=idx, word=all_words[idx]): + assert torch.sum(word_i == word_j).eq(min_freq_embed.embed_size) \ No newline at end of file diff --git a/test/io/loader/test_classification_loader.py b/test/io/loader/test_classification_loader.py new file mode 100644 index 00000000..f099c1b2 --- /dev/null +++ b/test/io/loader/test_classification_loader.py @@ -0,0 +1,27 @@ + +import unittest +from fastNLP.io.loader.classification import YelpFullLoader +from fastNLP.io.loader.classification import YelpPolarityLoader +from fastNLP.io.loader.classification import IMDBLoader +from fastNLP.io.loader.classification import SST2Loader +from fastNLP.io.loader.classification import SSTLoader +from fastNLP.io.loader.classification import ChnSentiCorpLoader +import os + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestDownload(unittest.TestCase): + def test_download(self): + for loader in [YelpFullLoader, YelpPolarityLoader, IMDBLoader, SST2Loader, SSTLoader, ChnSentiCorpLoader]: + loader().download() + + def test_load(self): + for loader in [YelpFullLoader, YelpPolarityLoader, IMDBLoader, SST2Loader, SSTLoader, ChnSentiCorpLoader]: + data_bundle = loader().load() + print(data_bundle) + + +class TestLoad(unittest.TestCase): + def test_load(self): + for loader in [IMDBLoader]: + data_bundle = loader().load('test/data_for_tests/io/imdb') + print(data_bundle) diff --git a/test/io/loader/test_conll_loader.py b/test/io/loader/test_conll_loader.py new file mode 100644 index 00000000..31859a6b --- /dev/null +++ b/test/io/loader/test_conll_loader.py @@ -0,0 +1,31 @@ + +import unittest +import os +from fastNLP.io.loader.conll import MsraNERLoader, PeopleDailyNERLoader, WeiboNERLoader, \ + Conll2003Loader + + +class TestMSRANER(unittest.TestCase): + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_download(self): + MsraNERLoader().download(re_download=False) + data_bundle = MsraNERLoader().load() + print(data_bundle) + + +class TestPeopleDaily(unittest.TestCase): + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_download(self): + PeopleDailyNERLoader().download() + + +class TestWeiboNER(unittest.TestCase): + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_download(self): + WeiboNERLoader().download() + + +class TestConll2003Loader(unittest.TestCase): + def test__load(self): + Conll2003Loader()._load('test/data_for_tests/conll_2003_example.txt') + diff --git a/test/io/loader/test_cws_loader.py b/test/io/loader/test_cws_loader.py new file mode 100644 index 00000000..55e48910 --- /dev/null +++ b/test/io/loader/test_cws_loader.py @@ -0,0 +1,24 @@ +import unittest +import os +from fastNLP.io.loader import CWSLoader + + +class TestCWSLoader(unittest.TestCase): + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_download(self): + dataset_names = ['pku', 'cityu', 'as', 'msra'] + for dataset_name in dataset_names: + with self.subTest(dataset_name=dataset_name): + data_bundle = CWSLoader(dataset_name=dataset_name).load() + print(data_bundle) + + +class TestRunCWSLoader(unittest.TestCase): + def test_cws_loader(self): + dataset_names = ['msra'] + for dataset_name in dataset_names: + with self.subTest(dataset_name=dataset_name): + data_bundle = CWSLoader(dataset_name=dataset_name).load( + f'test/data_for_tests/io/cws_{dataset_name}' + ) + print(data_bundle) diff --git a/test/io/loader/test_matching_loader.py b/test/io/loader/test_matching_loader.py new file mode 100644 index 00000000..cb1334e0 --- /dev/null +++ b/test/io/loader/test_matching_loader.py @@ -0,0 +1,29 @@ + +import unittest +from fastNLP.io.loader.matching import RTELoader +from fastNLP.io.loader.matching import QNLILoader +from fastNLP.io.loader.matching import SNLILoader +from fastNLP.io.loader.matching import QuoraLoader +from fastNLP.io.loader.matching import MNLILoader +import os + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestMatchingDownload(unittest.TestCase): + def test_download(self): + for loader in [RTELoader, QNLILoader, SNLILoader, MNLILoader]: + loader().download() + with self.assertRaises(Exception): + QuoraLoader().load() + + def test_load(self): + for loader in [RTELoader, QNLILoader, SNLILoader, MNLILoader]: + data_bundle = loader().load() + print(data_bundle) + + +class TestMatchingLoad(unittest.TestCase): + def test_load(self): + for loader in [RTELoader]: + data_bundle = loader().load('test/data_for_tests/io/rte') + print(data_bundle) + diff --git a/test/io/pipe/test_classification.py b/test/io/pipe/test_classification.py new file mode 100644 index 00000000..45c276a3 --- /dev/null +++ b/test/io/pipe/test_classification.py @@ -0,0 +1,30 @@ +import unittest +import os + +from fastNLP.io.pipe.classification import SSTPipe, SST2Pipe, IMDBPipe, YelpFullPipe, YelpPolarityPipe +from fastNLP.io.pipe.classification import ChnSentiCorpPipe + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestClassificationPipe(unittest.TestCase): + def test_process_from_file(self): + for pipe in [YelpPolarityPipe, SST2Pipe, IMDBPipe, YelpFullPipe, SSTPipe]: + with self.subTest(pipe=pipe): + print(pipe) + data_bundle = pipe(tokenizer='raw').process_from_file() + print(data_bundle) + + +class TestRunPipe(unittest.TestCase): + def test_load(self): + for pipe in [IMDBPipe]: + data_bundle = pipe(tokenizer='raw').process_from_file('test/data_for_tests/io/imdb') + print(data_bundle) + + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestCNClassificationPipe(unittest.TestCase): + def test_process_from_file(self): + for pipe in [ChnSentiCorpPipe]: + with self.subTest(pipe=pipe): + data_bundle = pipe(bigrams=True, trigrams=True).process_from_file() + print(data_bundle) \ No newline at end of file diff --git a/test/io/pipe/test_conll.py b/test/io/pipe/test_conll.py new file mode 100644 index 00000000..4ecd7969 --- /dev/null +++ b/test/io/pipe/test_conll.py @@ -0,0 +1,24 @@ +import unittest +import os +from fastNLP.io import MsraNERPipe, PeopleDailyPipe, WeiboNERPipe, Conll2003Pipe, Conll2003NERPipe + + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestConllPipe(unittest.TestCase): + def test_process_from_file(self): + for pipe in [MsraNERPipe, PeopleDailyPipe, WeiboNERPipe]: + with self.subTest(pipe=pipe): + print(pipe) + data_bundle = pipe(bigrams=True, trigrams=True).process_from_file() + print(data_bundle) + data_bundle = pipe(encoding_type='bioes').process_from_file() + print(data_bundle) + + +class TestRunPipe(unittest.TestCase): + def test_conll2003(self): + for pipe in [Conll2003Pipe, Conll2003NERPipe]: + with self.subTest(pipe=pipe): + print(pipe) + data_bundle = pipe().process_from_file('test/data_for_tests/conll_2003_example.txt') + print(data_bundle) diff --git a/test/io/pipe/test_cws.py b/test/io/pipe/test_cws.py new file mode 100644 index 00000000..063b6d9a --- /dev/null +++ b/test/io/pipe/test_cws.py @@ -0,0 +1,23 @@ + +import unittest +import os +from fastNLP.io.pipe.cws import CWSPipe + + +class TestCWSPipe(unittest.TestCase): + @unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") + def test_process_from_file(self): + dataset_names = ['pku', 'cityu', 'as', 'msra'] + for dataset_name in dataset_names: + with self.subTest(dataset_name=dataset_name): + data_bundle = CWSPipe(dataset_name=dataset_name).process_from_file() + print(data_bundle) + + +class TestRunCWSPipe(unittest.TestCase): + def test_process_from_file(self): + dataset_names = ['msra'] + for dataset_name in dataset_names: + with self.subTest(dataset_name=dataset_name): + data_bundle = CWSPipe().process_from_file(f'test/data_for_tests/io/cws_{dataset_name}') + print(data_bundle) diff --git a/test/io/pipe/test_matching.py b/test/io/pipe/test_matching.py new file mode 100644 index 00000000..932d8289 --- /dev/null +++ b/test/io/pipe/test_matching.py @@ -0,0 +1,34 @@ + +import unittest +import os + +from fastNLP.io.pipe.matching import SNLIPipe, RTEPipe, QNLIPipe, MNLIPipe +from fastNLP.io.pipe.matching import SNLIBertPipe, RTEBertPipe, QNLIBertPipe, MNLIBertPipe + + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestMatchingPipe(unittest.TestCase): + def test_process_from_file(self): + for pipe in [SNLIPipe, RTEPipe, QNLIPipe, MNLIPipe]: + with self.subTest(pipe=pipe): + print(pipe) + data_bundle = pipe(tokenizer='raw').process_from_file() + print(data_bundle) + + +@unittest.skipIf('TRAVIS' in os.environ, "Skip in travis") +class TestMatchingBertPipe(unittest.TestCase): + def test_process_from_file(self): + for pipe in [SNLIBertPipe, RTEBertPipe, QNLIBertPipe, MNLIBertPipe]: + with self.subTest(pipe=pipe): + print(pipe) + data_bundle = pipe(tokenizer='raw').process_from_file() + print(data_bundle) + + +class TestRunMatchingPipe(unittest.TestCase): + + def test_load(self): + for pipe in [RTEPipe, RTEBertPipe]: + data_bundle = pipe(tokenizer='raw').process_from_file('test/data_for_tests/io/rte') + print(data_bundle) diff --git a/test/io/test_dataset_loader.py b/test/io/test_dataset_loader.py deleted file mode 100644 index 492545f6..00000000 --- a/test/io/test_dataset_loader.py +++ /dev/null @@ -1,77 +0,0 @@ -import unittest -import os -from fastNLP.io import CSVLoader, JsonLoader -from fastNLP.io.data_loader import SSTLoader, SNLILoader, Conll2003Loader, PeopleDailyCorpusLoader - - -class TestDatasetLoader(unittest.TestCase): - - def test_Conll2003Loader(self): - """ - Test the the loader of Conll2003 dataset - """ - dataset_path = "test/data_for_tests/conll_2003_example.txt" - loader = Conll2003Loader() - dataset_2003 = loader.load(dataset_path) - - def test_PeopleDailyCorpusLoader(self): - data_set = PeopleDailyCorpusLoader().load("test/data_for_tests/people_daily_raw.txt") - - def test_CSVLoader(self): - ds = CSVLoader(sep='\t', headers=['words', 'label']) \ - .load('test/data_for_tests/tutorial_sample_dataset.csv') - assert len(ds) > 0 - - def test_SNLILoader(self): - ds = SNLILoader().load('test/data_for_tests/sample_snli.jsonl') - assert len(ds) == 3 - - def test_JsonLoader(self): - ds = JsonLoader().load('test/data_for_tests/sample_snli.jsonl') - assert len(ds) == 3 - - def no_test_SST(self): - train_data = """(3 (2 (2 The) (2 Rock)) (4 (3 (2 is) (4 (2 destined) (2 (2 (2 (2 (2 to) (2 (2 be) (2 (2 the) (2 (2 21st) (2 (2 (2 Century) (2 's)) (2 (3 new) (2 (2 ``) (2 Conan)))))))) (2 '')) (2 and)) (3 (2 that) (3 (2 he) (3 (2 's) (3 (2 going) (3 (2 to) (4 (3 (2 make) (3 (3 (2 a) (3 splash)) (2 (2 even) (3 greater)))) (2 (2 than) (2 (2 (2 (2 (1 (2 Arnold) (2 Schwarzenegger)) (2 ,)) (2 (2 Jean-Claud) (2 (2 Van) (2 Damme)))) (2 or)) (2 (2 Steven) (2 Segal))))))))))))) (2 .))) -(4 (4 (4 (2 The) (4 (3 gorgeously) (3 (2 elaborate) (2 continuation)))) (2 (2 (2 of) (2 ``)) (2 (2 The) (2 (2 (2 Lord) (2 (2 of) (2 (2 the) (2 Rings)))) (2 (2 '') (2 trilogy)))))) (2 (3 (2 (2 is) (2 (2 so) (2 huge))) (2 (2 that) (3 (2 (2 (2 a) (2 column)) (2 (2 of) (2 words))) (2 (2 (2 (2 can) (1 not)) (3 adequately)) (2 (2 describe) (2 (3 (2 (2 co-writer\/director) (2 (2 Peter) (3 (2 Jackson) (2 's)))) (3 (2 expanded) (2 vision))) (2 (2 of) (2 (2 (2 J.R.R.) (2 (2 Tolkien) (2 's))) (2 Middle-earth))))))))) (2 .))) -(3 (3 (2 (2 (2 (2 (2 Singer\/composer) (2 (2 Bryan) (2 Adams))) (2 (2 contributes) (2 (2 (2 a) (2 slew)) (2 (2 of) (2 songs))))) (2 (2 --) (2 (2 (2 (2 a) (2 (2 few) (3 potential))) (2 (2 (2 hits) (2 ,)) (2 (2 (2 a) (2 few)) (1 (1 (2 more) (1 (2 simply) (2 intrusive))) (2 (2 to) (2 (2 the) (2 story))))))) (2 --)))) (2 but)) (3 (4 (2 the) (3 (2 whole) (2 package))) (2 (3 certainly) (3 (2 captures) (2 (1 (2 the) (2 (2 (2 intended) (2 (2 ,) (2 (2 er) (2 ,)))) (3 spirit))) (2 (2 of) (2 (2 the) (2 piece)))))))) (2 .)) -(2 (2 (2 You) (2 (2 'd) (2 (2 think) (2 (2 by) (2 now))))) (2 (2 America) (2 (2 (2 would) (1 (2 have) (2 (2 (2 had) (1 (2 enough) (2 (2 of) (2 (2 plucky) (2 (2 British) (1 eccentrics)))))) (4 (2 with) (4 (3 hearts) (3 (2 of) (3 gold))))))) (2 .)))) -""" - test_data = """(3 (2 Yet) (3 (2 (2 the) (2 act)) (3 (4 (3 (2 is) (3 (2 still) (4 charming))) (2 here)) (2 .)))) -(4 (2 (2 Whether) (2 (2 (2 (2 or) (1 not)) (3 (2 you) (2 (2 're) (3 (3 enlightened) (2 (2 by) (2 (2 any) (2 (2 of) (2 (2 Derrida) (2 's))))))))) (2 (2 lectures) (2 (2 on) (2 (2 ``) (2 (2 (2 (2 (2 (2 the) (2 other)) (2 '')) (2 and)) (2 ``)) (2 (2 the) (2 self)))))))) (3 (2 ,) (3 (2 '') (3 (2 Derrida) (3 (3 (2 is) (4 (2 an) (4 (4 (2 undeniably) (3 (4 (3 fascinating) (2 and)) (4 playful))) (2 fellow)))) (2 .)))))) -(4 (3 (2 (2 Just) (2 (2 the) (2 labour))) (3 (2 involved) (3 (2 in) (4 (2 creating) (3 (3 (2 the) (3 (3 layered) (2 richness))) (3 (2 of) (3 (2 (2 the) (2 imagery)) (2 (2 in) (3 (2 (2 this) (2 chiaroscuro)) (2 (2 of) (2 (2 (2 madness) (2 and)) (2 light)))))))))))) (3 (3 (2 is) (4 astonishing)) (2 .))) -(3 (3 (2 Part) (3 (2 of) (4 (2 (2 the) (3 charm)) (2 (2 of) (2 (2 Satin) (2 Rouge)))))) (3 (3 (2 is) (3 (2 that) (3 (2 it) (2 (1 (2 avoids) (2 (2 the) (1 obvious))) (3 (2 with) (3 (3 (3 humour) (2 and)) (2 lightness))))))) (2 .))) -(4 (2 (2 a) (2 (2 screenplay) (2 more))) (3 (4 ingeniously) (2 (2 constructed) (2 (2 (2 (2 than) (2 ``)) (2 Memento)) (2 ''))))) -(3 (2 ``) (3 (2 (2 Extreme) (2 Ops)) (3 (2 '') (4 (4 (3 exceeds) (2 expectations)) (2 .))))) -""" - train, test = 'train--', 'test--' - with open(train, 'w', encoding='utf-8') as f: - f.write(train_data) - with open(test, 'w', encoding='utf-8') as f: - f.write(test_data) - - loader = SSTLoader() - info = loader.process( - {train: train, test: test}, - train_ds=[train], - src_vocab_op=dict(min_freq=2) - ) - assert len(list(info.vocabs.items())) == 2 - assert len(list(info.datasets.items())) == 2 - print(info.vocabs) - print(info.datasets) - os.remove(train), os.remove(test) - - def test_import(self): - import fastNLP - from fastNLP.io import SNLILoader - ds = SNLILoader().process('test/data_for_tests/sample_snli.jsonl', to_lower=True, - get_index=True, seq_len_type='seq_len', extra_split=['-']) - assert 'train' in ds.datasets - assert len(ds.datasets) == 1 - assert len(ds.datasets['train']) == 3 - - ds = SNLILoader().process('test/data_for_tests/sample_snli.jsonl', to_lower=True, - get_index=True, seq_len_type='seq_len') - assert 'train' in ds.datasets - assert len(ds.datasets) == 1 - assert len(ds.datasets['train']) == 3 diff --git a/test/io/test_embed_loader.py b/test/io/test_embed_loader.py index bbfe8858..70b367ec 100644 --- a/test/io/test_embed_loader.py +++ b/test/io/test_embed_loader.py @@ -8,8 +8,8 @@ from fastNLP.io import EmbedLoader class TestEmbedLoader(unittest.TestCase): def test_load_with_vocab(self): vocab = Vocabulary() - glove = "test/data_for_tests/glove.6B.50d_test.txt" - word2vec = "test/data_for_tests/word2vec_test.txt" + glove = "test/data_for_tests/embedding/small_static_embedding/glove.6B.50d_test.txt" + word2vec = "test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt" vocab.add_word('the') vocab.add_word('none') g_m = EmbedLoader.load_with_vocab(glove, vocab) @@ -20,8 +20,8 @@ class TestEmbedLoader(unittest.TestCase): def test_load_without_vocab(self): words = ['the', 'of', 'in', 'a', 'to', 'and'] - glove = "test/data_for_tests/glove.6B.50d_test.txt" - word2vec = "test/data_for_tests/word2vec_test.txt" + glove = "test/data_for_tests/embedding/small_static_embedding/glove.6B.50d_test.txt" + word2vec = "test/data_for_tests/embedding/small_static_embedding/word2vec_test.txt" g_m, vocab = EmbedLoader.load_without_vocab(glove) self.assertEqual(g_m.shape, (8, 50)) for word in words: diff --git a/test/models/test_bert.py b/test/models/test_bert.py index 05ee6d5a..9cab3a88 100644 --- a/test/models/test_bert.py +++ b/test/models/test_bert.py @@ -2,68 +2,174 @@ import unittest import torch -from fastNLP.models.bert import * +from fastNLP.core import Vocabulary, Const +from fastNLP.models.bert import BertForSequenceClassification, BertForQuestionAnswering, \ + BertForTokenClassification, BertForMultipleChoice, BertForSentenceMatching +from fastNLP.embeddings.bert_embedding import BertEmbedding class TestBert(unittest.TestCase): def test_bert_1(self): - from fastNLP.core.const import Const - from fastNLP.modules.encoder.bert import BertConfig + vocab = Vocabulary().add_word_lst("this is a test .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=True) - model = BertForSequenceClassification(2, BertConfig(32000)) + model = BertForSequenceClassification(embed, 2) - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) + input_ids = torch.LongTensor([[1, 2, 3], [5, 6, 0]]) - pred = model(input_ids, token_type_ids, input_mask) + pred = model(input_ids) self.assertTrue(isinstance(pred, dict)) self.assertTrue(Const.OUTPUT in pred) self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2, 2)) + pred = model(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUT in pred) + self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2, 2)) + + def test_bert_1_w(self): + vocab = Vocabulary().add_word_lst("this is a test .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=False) + + with self.assertWarns(Warning): + model = BertForSequenceClassification(embed, 2) + + input_ids = torch.LongTensor([[1, 2, 3], [5, 6, 0]]) + + pred = model.predict(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUT in pred) + self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2,)) + def test_bert_2(self): - from fastNLP.core.const import Const - from fastNLP.modules.encoder.bert import BertConfig - model = BertForMultipleChoice(2, BertConfig(32000)) + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=True) - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) + model = BertForMultipleChoice(embed, 2) - pred = model(input_ids, token_type_ids, input_mask) + input_ids = torch.LongTensor([[[2, 6, 7], [1, 6, 5]]]) + print(input_ids.size()) + + pred = model(input_ids) self.assertTrue(isinstance(pred, dict)) self.assertTrue(Const.OUTPUT in pred) self.assertEqual(tuple(pred[Const.OUTPUT].shape), (1, 2)) + def test_bert_2_w(self): + + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=False) + + with self.assertWarns(Warning): + model = BertForMultipleChoice(embed, 2) + + input_ids = torch.LongTensor([[[2, 6, 7], [1, 6, 5]]]) + print(input_ids.size()) + + pred = model.predict(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUT in pred) + self.assertEqual(tuple(pred[Const.OUTPUT].shape), (1,)) + def test_bert_3(self): - from fastNLP.core.const import Const - from fastNLP.modules.encoder.bert import BertConfig - model = BertForTokenClassification(7, BertConfig(32000)) + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=False) + model = BertForTokenClassification(embed, 7) - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) + input_ids = torch.LongTensor([[1, 2, 3], [6, 5, 0]]) - pred = model(input_ids, token_type_ids, input_mask) + pred = model(input_ids) self.assertTrue(isinstance(pred, dict)) self.assertTrue(Const.OUTPUT in pred) self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2, 3, 7)) + def test_bert_3_w(self): + + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=True) + + with self.assertWarns(Warning): + model = BertForTokenClassification(embed, 7) + + input_ids = torch.LongTensor([[1, 2, 3], [6, 5, 0]]) + + pred = model.predict(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUT in pred) + self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2, 3)) + def test_bert_4(self): - from fastNLP.core.const import Const - from fastNLP.modules.encoder.bert import BertConfig - model = BertForQuestionAnswering(BertConfig(32000)) + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=True) + model = BertForQuestionAnswering(embed) - input_ids = torch.LongTensor([[31, 51, 99], [15, 5, 0]]) - input_mask = torch.LongTensor([[1, 1, 1], [1, 1, 0]]) - token_type_ids = torch.LongTensor([[0, 0, 1], [0, 1, 0]]) + input_ids = torch.LongTensor([[1, 2, 3], [6, 5, 0]]) - pred = model(input_ids, token_type_ids, input_mask) + pred = model(input_ids) self.assertTrue(isinstance(pred, dict)) self.assertTrue(Const.OUTPUTS(0) in pred) self.assertTrue(Const.OUTPUTS(1) in pred) - self.assertEqual(tuple(pred[Const.OUTPUTS(0)].shape), (2, 3)) - self.assertEqual(tuple(pred[Const.OUTPUTS(1)].shape), (2, 3)) + self.assertEqual(tuple(pred[Const.OUTPUTS(0)].shape), (2, 5)) + self.assertEqual(tuple(pred[Const.OUTPUTS(1)].shape), (2, 5)) + + model = BertForQuestionAnswering(embed, 7) + pred = model(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertEqual(len(pred), 7) + + def test_bert_4_w(self): + + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=False) + + with self.assertWarns(Warning): + model = BertForQuestionAnswering(embed) + + input_ids = torch.LongTensor([[1, 2, 3], [6, 5, 0]]) + + pred = model.predict(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUTS(1) in pred) + self.assertEqual(tuple(pred[Const.OUTPUTS(1)].shape), (2,)) + + def test_bert_5(self): + + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=True) + model = BertForSentenceMatching(embed) + + input_ids = torch.LongTensor([[1, 2, 3], [6, 5, 0]]) + + pred = model(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUT in pred) + self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2, 2)) + + def test_bert_5_w(self): + + vocab = Vocabulary().add_word_lst("this is a test [SEP] .".split()) + embed = BertEmbedding(vocab, model_dir_or_name='test/data_for_tests/embedding/small_bert', + include_cls_sep=False) + + with self.assertWarns(Warning): + model = BertForSentenceMatching(embed) + + input_ids = torch.LongTensor([[1, 2, 3], [6, 5, 0]]) + + pred = model.predict(input_ids) + self.assertTrue(isinstance(pred, dict)) + self.assertTrue(Const.OUTPUT in pred) + self.assertEqual(tuple(pred[Const.OUTPUT].shape), (2,)) + diff --git a/test/models/test_biaffine_parser.py b/test/models/test_biaffine_parser.py index 4f93b994..4b38d816 100644 --- a/test/models/test_biaffine_parser.py +++ b/test/models/test_biaffine_parser.py @@ -27,7 +27,7 @@ def prepare_parser_data(): class TestBiaffineParser(unittest.TestCase): def test_train(self): - model = BiaffineParser(init_embed=(VOCAB_SIZE, 10), + model = BiaffineParser(embed=(VOCAB_SIZE, 10), pos_vocab_size=VOCAB_SIZE, pos_emb_dim=10, rnn_hidden_size=10, arc_mlp_size=10, @@ -37,7 +37,7 @@ class TestBiaffineParser(unittest.TestCase): RUNNER.run_model(model, ds, loss=ParserLoss(), metrics=ParserMetric()) def test_train2(self): - model = BiaffineParser(init_embed=(VOCAB_SIZE, 10), + model = BiaffineParser(embed=(VOCAB_SIZE, 10), pos_vocab_size=VOCAB_SIZE, pos_emb_dim=10, rnn_hidden_size=16, arc_mlp_size=10, diff --git a/test/models/test_sequence_labeling.py b/test/models/test_sequence_labeling.py index 3a70e381..815d7047 100644 --- a/test/models/test_sequence_labeling.py +++ b/test/models/test_sequence_labeling.py @@ -3,9 +3,24 @@ import unittest from .model_runner import * -from fastNLP.models.sequence_labeling import SeqLabeling, AdvSeqLabel +from fastNLP.models.sequence_labeling import SeqLabeling, AdvSeqLabel, BiLSTMCRF from fastNLP.core.losses import LossInForward +class TestBiLSTM(unittest.TestCase): + def test_case1(self): + # 测试能否正常运行CNN + init_emb = (VOCAB_SIZE, 30) + model = BiLSTMCRF(init_emb, + hidden_size=30, + num_classes=NUM_CLS) + + data = RUNNER.prepare_pos_tagging_data() + data.set_input('target') + loss = LossInForward() + metric = AccuracyMetric(pred=C.OUTPUT, target=C.TARGET, seq_len=C.INPUT_LEN) + RUNNER.run_model(model, data, loss, metric) + + class TesSeqLabel(unittest.TestCase): def test_case1(self): # 测试能否正常运行CNN diff --git a/test/models/test_snli.py b/test/models/test_snli.py new file mode 100644 index 00000000..7a588a4c --- /dev/null +++ b/test/models/test_snli.py @@ -0,0 +1,9 @@ +import unittest +from .model_runner import * +from fastNLP.models.snli import ESIM + + +class TestSNLIModel(unittest.TestCase): + def test_snli(self): + model = ESIM((VOCAB_SIZE, 10), num_labels=NUM_CLS, dropout_rate=0) + RUNNER.run_model_with_task(NLI, model) diff --git a/test/modules/decoder/test_CRF.py b/test/modules/decoder/test_CRF.py index 647af7d3..94b4ab7a 100644 --- a/test/modules/decoder/test_CRF.py +++ b/test/modules/decoder/test_CRF.py @@ -1,6 +1,6 @@ import unittest - +from fastNLP import Vocabulary class TestCRF(unittest.TestCase): def test_case1(self): @@ -14,7 +14,8 @@ class TestCRF(unittest.TestCase): id2label = {0: 'B', 1:'M', 2:'E', 3:'S'} expected_res = {(0, 1), (0, 2), (1, 1), (1, 2), (2, 0), (2, 3), (2, 5), (3, 0), (3, 3), (3, 5), (4, 0), (4, 3)} - self.assertSetEqual(expected_res, set(allowed_transitions(id2label, encoding_type='BMES', include_start_end=True))) + self.assertSetEqual(expected_res, set( + allowed_transitions(id2label, encoding_type='BMES', include_start_end=True))) id2label = {0: 'B', 1: 'I', 2:'O', 3: '', 4:""} allowed_transitions(id2label, include_start_end=True) @@ -37,7 +38,100 @@ class TestCRF(unittest.TestCase): expected_res = {(0, 1), (0, 2), (1, 1), (1, 2), (2, 0), (2, 3), (2, 4), (2, 7), (2, 9), (3, 0), (3, 3), (3, 4), (3, 7), (3, 9), (4, 5), (4, 6), (5, 5), (5, 6), (6, 0), (6, 3), (6, 4), (6, 7), (6, 9), (7, 0), (7, 3), (7, 4), (7, 7), (7, 9), (8, 0), (8, 3), (8, 4), (8, 7)} - self.assertSetEqual(expected_res, set(allowed_transitions(id2label, encoding_type='BMES', include_start_end=True))) + self.assertSetEqual(expected_res, set( + allowed_transitions(id2label, include_start_end=True))) + + def test_case11(self): + # 测试自动推断encoding类型 + from fastNLP.modules.decoder.crf import allowed_transitions + + id2label = {0: 'B', 1: 'I', 2: 'O'} + expected_res = {(0, 0), (0, 1), (0, 2), (0, 4), (1, 0), (1, 1), (1, 2), (1, 4), (2, 0), (2, 2), + (2, 4), (3, 0), (3, 2)} + self.assertSetEqual(expected_res, set(allowed_transitions(id2label, include_start_end=True))) + + id2label = {0: 'B', 1: 'M', 2: 'E', 3: 'S'} + expected_res = {(0, 1), (0, 2), (1, 1), (1, 2), (2, 0), (2, 3), (2, 5), (3, 0), (3, 3), (3, 5), (4, 0), (4, 3)} + self.assertSetEqual(expected_res, set( + allowed_transitions(id2label, include_start_end=True))) + + id2label = {0: 'B', 1: 'I', 2: 'O', 3: '', 4: ""} + allowed_transitions(id2label, include_start_end=True) + + labels = ['O'] + for label in ['X', 'Y']: + for tag in 'BI': + labels.append('{}-{}'.format(tag, label)) + id2label = {idx: label for idx, label in enumerate(labels)} + expected_res = {(0, 0), (0, 1), (0, 3), (0, 6), (1, 0), (1, 1), (1, 2), (1, 3), (1, 6), (2, 0), (2, 1), + (2, 2), (2, 3), (2, 6), (3, 0), (3, 1), (3, 3), (3, 4), (3, 6), (4, 0), (4, 1), (4, 3), + (4, 4), (4, 6), (5, 0), (5, 1), (5, 3)} + self.assertSetEqual(expected_res, set(allowed_transitions(id2label, include_start_end=True))) + + labels = [] + for label in ['X', 'Y']: + for tag in 'BMES': + labels.append('{}-{}'.format(tag, label)) + id2label = {idx: label for idx, label in enumerate(labels)} + expected_res = {(0, 1), (0, 2), (1, 1), (1, 2), (2, 0), (2, 3), (2, 4), (2, 7), (2, 9), (3, 0), (3, 3), (3, 4), + (3, 7), (3, 9), (4, 5), (4, 6), (5, 5), (5, 6), (6, 0), (6, 3), (6, 4), (6, 7), (6, 9), (7, 0), + (7, 3), (7, 4), (7, 7), (7, 9), (8, 0), (8, 3), (8, 4), (8, 7)} + self.assertSetEqual(expected_res, set( + allowed_transitions(id2label, include_start_end=True))) + + def test_case12(self): + # 测试能否通过vocab生成转移矩阵 + from fastNLP.modules.decoder.crf import allowed_transitions + + id2label = {0: 'B', 1: 'I', 2: 'O'} + vocab = Vocabulary(unknown=None, padding=None) + for idx, tag in id2label.items(): + vocab.add_word(tag) + expected_res = {(0, 0), (0, 1), (0, 2), (0, 4), (1, 0), (1, 1), (1, 2), (1, 4), (2, 0), (2, 2), + (2, 4), (3, 0), (3, 2)} + self.assertSetEqual(expected_res, set(allowed_transitions(vocab, include_start_end=True))) + + id2label = {0: 'B', 1: 'M', 2: 'E', 3: 'S'} + vocab = Vocabulary(unknown=None, padding=None) + for idx, tag in id2label.items(): + vocab.add_word(tag) + expected_res = {(0, 1), (0, 2), (1, 1), (1, 2), (2, 0), (2, 3), (2, 5), (3, 0), (3, 3), (3, 5), (4, 0), (4, 3)} + self.assertSetEqual(expected_res, set( + allowed_transitions(vocab, include_start_end=True))) + + id2label = {0: 'B', 1: 'I', 2: 'O', 3: '', 4: ""} + vocab = Vocabulary() + for idx, tag in id2label.items(): + vocab.add_word(tag) + allowed_transitions(vocab, include_start_end=True) + + labels = ['O'] + for label in ['X', 'Y']: + for tag in 'BI': + labels.append('{}-{}'.format(tag, label)) + id2label = {idx: label for idx, label in enumerate(labels)} + expected_res = {(0, 0), (0, 1), (0, 3), (0, 6), (1, 0), (1, 1), (1, 2), (1, 3), (1, 6), (2, 0), (2, 1), + (2, 2), (2, 3), (2, 6), (3, 0), (3, 1), (3, 3), (3, 4), (3, 6), (4, 0), (4, 1), (4, 3), + (4, 4), (4, 6), (5, 0), (5, 1), (5, 3)} + vocab = Vocabulary(unknown=None, padding=None) + for idx, tag in id2label.items(): + vocab.add_word(tag) + self.assertSetEqual(expected_res, set(allowed_transitions(vocab, include_start_end=True))) + + labels = [] + for label in ['X', 'Y']: + for tag in 'BMES': + labels.append('{}-{}'.format(tag, label)) + id2label = {idx: label for idx, label in enumerate(labels)} + vocab = Vocabulary(unknown=None, padding=None) + for idx, tag in id2label.items(): + vocab.add_word(tag) + expected_res = {(0, 1), (0, 2), (1, 1), (1, 2), (2, 0), (2, 3), (2, 4), (2, 7), (2, 9), (3, 0), (3, 3), (3, 4), + (3, 7), (3, 9), (4, 5), (4, 6), (5, 5), (5, 6), (6, 0), (6, 3), (6, 4), (6, 7), (6, 9), (7, 0), + (7, 3), (7, 4), (7, 7), (7, 9), (8, 0), (8, 3), (8, 4), (8, 7)} + self.assertSetEqual(expected_res, set( + allowed_transitions(vocab, include_start_end=True))) + def test_case2(self): # 测试CRF能否避免解码出非法跃迁, 使用allennlp做了验证。 diff --git a/test/modules/encoder/test_bert.py b/test/modules/decoder/test_bert.py similarity index 92% rename from test/modules/encoder/test_bert.py rename to test/modules/decoder/test_bert.py index 0fcf01e4..56946f5d 100644 --- a/test/modules/encoder/test_bert.py +++ b/test/modules/decoder/test_bert.py @@ -3,7 +3,7 @@ import unittest import torch -from fastNLP.models.bert import BertModel +from fastNLP.modules.encoder.bert import BertModel class TestBert(unittest.TestCase): diff --git a/test/test_tutorials.py b/test/test_tutorials.py index 6f4a8347..3ec0e381 100644 --- a/test/test_tutorials.py +++ b/test/test_tutorials.py @@ -5,14 +5,13 @@ from fastNLP import Instance from fastNLP import Vocabulary from fastNLP.core.losses import CrossEntropyLoss from fastNLP.core.metrics import AccuracyMetric - +from fastNLP.io.loader import CSVLoader class TestTutorial(unittest.TestCase): def test_fastnlp_10min_tutorial(self): # 从csv读取数据到DataSet sample_path = "test/data_for_tests/tutorial_sample_dataset.csv" - dataset = DataSet.read_csv(sample_path, headers=('raw_sentence', 'label'), - sep='\t') + dataset = CSVLoader(headers=['raw_sentence', 'label'], sep=' ')._load(sample_path) print(len(dataset)) print(dataset[0]) print(dataset[-3]) @@ -110,7 +109,7 @@ class TestTutorial(unittest.TestCase): def test_fastnlp_1min_tutorial(self): # tutorials/fastnlp_1min_tutorial.ipynb data_path = "test/data_for_tests/tutorial_sample_dataset.csv" - ds = DataSet.read_csv(data_path, headers=('raw_sentence', 'label'), sep='\t') + ds = CSVLoader(headers=['raw_sentence', 'label'], sep=' ')._load(data_path) print(ds[1]) # 将所有数字转为小写