You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

readme.md 2.4 kB

2 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283
  1. # train stage
  2. `python darts_train.py --data_dir '../data/' --result_path 'trial_id/result.json' --log_path 'trial_id/log' --search_space_path 'experiment_id/search_space.json' --best_selected_space_path 'experiment_id/best_selected_space.json' --trial_id 0 --layers 8 --model_lr 0.025 --arch_lr 3e-4 --epochs 1 --batch_size 64 --channels 16`
  3. Note:
  4. here `--epochs 2` just for debug
  5. # select stage
  6. `python darts_select.py --best_selected_space_path 'experiment_id/best_selected_space.json' `
  7. # retrain stage
  8. `python darts_retrain.py --data_dir '../data/' --result_path 'trial_id/result.json' --log_path 'trial_id/log' --best_selected_space_path 'experiment_id/best_selected_space.json' --best_checkpoint_dir 'experiment_id/' --trial_id 0 --batch_size 96 --epochs 1 --lr 0.025 --layers 20 --channels 36`
  9. # output file
  10. `result.json`
  11. ```
  12. {'type': 'Accuracy', 'result': {'sequence': 0, 'category': 'epoch', 'value': 0.1}}
  13. {'type': 'Accuracy', 'result': {'sequence': 1, 'category': 'epoch', 'value': 0.0}}
  14. {'type': 'Accuracy', 'result': {'sequence': 2, 'category': 'epoch', 'value': 0.0}}
  15. {'type': 'Accuracy', 'result': {'sequence': 3, 'category': 'epoch', 'value': 0.0}}
  16. {'type': 'Accuracy', 'result': {'sequence': 4, 'category': 'epoch', 'value': 0.0}}
  17. {'type': 'Cost_time', 'result': {'value': '41.614346981048584 s'}}
  18. ```
  19. `search_space.json`
  20. ```
  21. {
  22. "op_list": {
  23. "_type": "layer_choice",
  24. "_value": [
  25. "maxpool",
  26. "avgpool",
  27. "skipconnect",
  28. "sepconv3x3",
  29. "sepconv5x5",
  30. "dilconv3x3",
  31. "dilconv5x5",
  32. "none"
  33. ]
  34. },
  35. "search_space": {
  36. "normal_n2_p0": "op_list",
  37. "normal_n2_p1": "op_list",
  38. "normal_n2_switch": {
  39. "_type": "input_choice",
  40. "_value": {
  41. "candidates": [
  42. "normal_n2_p0",
  43. "normal_n2_p1"
  44. ],
  45. "n_chosen": 2
  46. }
  47. },
  48. ...
  49. }
  50. ```
  51. `best_selected_space.json`
  52. ```
  53. {
  54. "normal_n2_p0": "dilconv5x5",
  55. "normal_n2_p1": "dilconv5x5",
  56. "normal_n2_switch": [
  57. "normal_n2_p0",
  58. "normal_n2_p1"
  59. ],
  60. "normal_n3_p0": "sepconv3x3",
  61. "normal_n3_p1": "dilconv5x5",
  62. "normal_n3_p2": [],
  63. "normal_n3_switch": [
  64. "normal_n3_p0",
  65. "normal_n3_p1"
  66. ],
  67. "normal_n4_p0": [],
  68. "normal_n4_p1": "dilconv5x5",
  69. "normal_n4_p2": "sepconv5x5",
  70. "normal_n4_p3": [],
  71. "normal_n4_switch": [
  72. "normal_n4_p1",
  73. "normal_n4_p2"
  74. ],
  75. ...
  76. }
  77. ```

一站式算法开发平台、高性能分布式深度学习框架、先进算法模型库、视觉模型炼知平台、数据可视化分析平台等一系列平台及工具,在模型高效分布式训练、数据处理和可视分析、模型炼知和轻量化等技术上形成独特优势,目前已在产学研等各领域近千家单位及个人提供AI应用赋能