diff --git a/model_inplement/code/evaluate.py b/model_inplement/code/evaluate.py
new file mode 100644
index 00000000..38e17e8a
--- /dev/null
+++ b/model_inplement/code/evaluate.py
@@ -0,0 +1,61 @@
+from model import *
+from train import *
+
+def evaluate(net, dataset, bactch_size=64, use_cuda=False):
+    dataloader = DataLoader(dataset, batch_size=bactch_size, collate_fn=collate, num_workers=0)
+    count = 0
+    if use_cuda:
+        net.cuda()
+    for i, batch_samples in enumerate(dataloader):
+        x, y = batch_samples
+        doc_list = []
+        for sample in x:
+            doc = []
+            for sent_vec in sample:
+                # print(sent_vec.size())
+                if use_cuda:
+                    sent_vec = sent_vec.cuda()
+                doc.append(Variable(sent_vec, volatile=True))
+            doc_list.append(pack_sequence(doc))
+        if use_cuda:
+            y = y.cuda()
+        predicts = net(doc_list)
+        # idx = []
+        # for p in predicts.data:
+        #     idx.append(np.random.choice(5, p=torch.exp(p).numpy()))
+        # idx = torch.LongTensor(idx)
+        p, idx = torch.max(predicts, dim=1)
+        idx = idx.data
+        count += torch.sum(torch.eq(idx, y))
+    return count
+
+def visualize_attention(doc, alpha_vec):
+    pass
+
+if __name__ == '__main__':
+    from gensim.models import Word2Vec
+    import gensim
+    from gensim import models
+    embed_model = Word2Vec.load('yelp.word2vec')
+    embedding = Embedding_layer(embed_model.wv, embed_model.wv.vector_size)
+    del embed_model
+
+    net = HAN(input_size=200, output_size=5, 
+            word_hidden_size=50, word_num_layers=1, word_context_size=100,
+            sent_hidden_size=50, sent_num_layers=1, sent_context_size=100)
+    net.load_state_dict(torch.load('model.dict'))
+    test_dataset = YelpDocSet('reviews', 199, 4, embedding)
+    correct = evaluate(net, test_dataset, True)
+    print('accuracy {}'.format(correct/len(test_dataset)))
+    # data_idx = 121
+    # x, y = test_dataset[data_idx]
+    # doc = []
+    # for sent_vec in x:
+    #     doc.append(Variable(sent_vec, volatile=True))
+    # input_vec = [pack_sequence(doc)]
+    # predict = net(input_vec)
+    # p, idx = torch.max(predict, dim=1)
+    # print(net.word_layer.last_alpha.squeeze())
+    # print(net.sent_layer.last_alpha)
+    # print(test_dataset.get_doc(data_idx)[0])
+    # print('predict: {}, true: {}'.format(int(idx), y))
diff --git a/model_inplement/code/model.py b/model_inplement/code/model.py
index 068eaa55..066a695b 100644
--- a/model_inplement/code/model.py
+++ b/model_inplement/code/model.py
@@ -55,6 +55,7 @@ class AttentionNet(nn.Module):
         self.gru_hidden_size = gru_hidden_size
         self.gru_num_layers = gru_num_layers
         self.context_vec_size = context_vec_size
+        self.last_alpha = None
 
         # Encoder
         self.gru = nn.GRU(input_size=input_size, 
@@ -76,6 +77,7 @@ class AttentionNet(nn.Module):
         u = self.tanh(self.fc(h_t))
         # u's dim (batch_size, seq_len, context_vec_size)
         alpha = self.softmax(torch.matmul(u, self.context_vec))
+        self.last_alpha = alpha.data
         # alpha's dim (batch_size, seq_len, 1)
         output = torch.bmm(torch.transpose(h_t, 1, 2), alpha)
         # output's dim (batch_size, 2*hidden_size, 1)
diff --git a/model_inplement/code/train.py b/model_inplement/code/train.py
index 1533a34a..d28ef8bb 100644
--- a/model_inplement/code/train.py
+++ b/model_inplement/code/train.py
@@ -78,6 +78,20 @@ class YelpDocSet(Dataset):
         self.embedding = embedding
         self._cache = [(-1, None) for i in range(5)]
 
+    def get_doc(self, n):
+        file_id = n // 5000
+        idx = file_id % 5
+        if self._cache[idx][0] != file_id:
+            print('load {} to {}'.format(file_id, idx))
+            with open(os.path.join(self.dirname, self._files[file_id]), 'rb') as f:
+                self._cache[idx] = (file_id, pickle.load(f))
+        y, x = self._cache[idx][1][n % 5000]
+        sents = []
+        for s_list in x:
+            sents.append(' '.join(s_list))
+        x = '\n'.join(sents)
+        return x, y-1
+
     def __len__(self):
         return len(self._files)*5000
 
@@ -166,7 +180,6 @@ if __name__ == '__main__':
     embed_model = Word2Vec.load('yelp.word2vec')
     embedding = Embedding_layer(embed_model.wv, embed_model.wv.vector_size)
     del embed_model
-    # for start_file in range(11, 24):
     start_file = 0
     dataset = YelpDocSet('reviews', start_file, 120-start_file, embedding)
     print('start_file %d'% start_file)
@@ -176,4 +189,4 @@ if __name__ == '__main__':
             sent_hidden_size=50, sent_num_layers=1, sent_context_size=100)
         
     net.load_state_dict(torch.load('model.dict'))
-    train(net, dataset, num_epoch=1, batch_size=64, use_cuda=True)
+    train(net, dataset, num_epoch=5, batch_size=64, use_cuda=True)
\ No newline at end of file