修改了transformer的log_max部分

Author: mathCrazyy

__pycache__/model.cpython-36.pyc

@@ -5,7 +5,7 @@
 from torchtext import data
 import torch.nn as nn
 import torch.nn.functional as F
-from train_eval import train,test
+from train_eval import train,test, test_one_sentence
 
 from torch.autograd import Variable as V
 
@@ -15,8 +15,8 @@
 class Config(object):
     def __init__(self):
         self.model_name="lm_model"
-        #self.data_ori="/mnt/data3/wuchunsheng/data_all/data_mine/lm_data/"
-        self.data_ori="E:/data/word_nlp/cnews_data/"
+        self.data_ori="/mnt/data3/wuchunsheng/data_all/data_mine/lm_data/"
+        #self.data_ori="E:/data/word_nlp/cnews_data/"
         self.train_path="train_0.csv"
         self.valid_path="train_0.csv"
         self.test_path="test_100.csv"
@@ -34,7 +34,7 @@ def __init__(self):
         self.hidden_size=200
         self.nlayers=1
         self.dropout=0.5
-        self.epoch=1
+        self.epoch=100
 
         self.train_len=0
         self.test_len = 0
@@ -59,6 +59,9 @@ def __init__(self):
 
 model=TransformerModel(config, TEXT).to(device)
 
-#train(config,model,train_iter, valid_iter, test_iter)
+train(config,model,train_iter, valid_iter, test_iter)
 
-test(config,model,TEXT,  test_iter)## 测试的是一个正批量的
+#res=test(config,model,TEXT,  test_iter)## 测试的是一个正批量的
+#print(res)
+res=test_one_sentence(config, model, TEXT, test_iter)
+print(res)

__pycache__/train_eval.cpython-36.pyc

@@ -16,8 +16,8 @@
 class Config(object):
     def __init__(self):
         self.model_name="lm_model"
-        #self.data_ori="/mnt/data3/wuchunsheng/data_all/data_mine/lm_data/"
-        self.data_ori="E:/data/word_nlp/cnews_data/"
+        self.data_ori="/mnt/data3/wuchunsheng/data_all/data_mine/lm_data/"
+        #self.data_ori="E:/data/word_nlp/cnews_data/"
         self.train_path="train_0.csv"
         self.valid_path="train_0.csv"
         self.test_path="test_100.csv"
@@ -35,7 +35,7 @@ def __init__(self):
         self.hidden_size=200
         self.nlayers=1
         self.dropout=0.5
-        self.epoch=2
+        self.epoch=20
 
         self.train_len=0
         self.test_len = 0
@@ -59,8 +59,24 @@ def __init__(self):
 
 model =load_model(config, model)
 
-sen="comment体育项目"
-sen="".join(['c', 'o', 'n', 't', 'e', 'x', 't', ',', 'l', 'a', 'b', 'e', 'l'])
-res=test_sentence(config, model ,TEXT, sen)
-print(res)
-
+#sen="目"*50
+sen="体育快讯"
+#sen="".join(['c', 'o', 'n', 't', 'e', 'x', 't', ',', 'l', 'a', 'b', 'e', 'l'])
+#res=test_sentence(config, model ,TEXT, sen)
+#print(sen)
+#print(res)
+#res=test(config,model,TEXT,  test_iter)
+#print(res)
+print("=========================")
+sen="篮球"
+#sen="体育"
+sen_ori=sen
+while(len(sen)<20):
+    print("输入文本: ",sen)
+    sen_pred=" ".join(test_sentence(config,model, TEXT,sen))
+    sen+=sen_pred[1:]
+    sen=sen.replace(" ","")
+    print("文本生成: ", sen)
+print("*"*20)
+print("输入: ", sen_ori)
+print("生成: ", sen)

__pycache__/transformer.cpython-36.pyc

@@ -7,7 +7,8 @@
 
 def train(config,model,train_iter, valid_iter,test_iter):
 
-    optimizer = optim.Adam(model.parameters(), lr=1e-3, betas=(0.7, 0.99))
+    #optimizer = optim.Adam(model.parameters(), lr=1e-3, betas=(0.7, 0.99))
+    optimizer = optim.Adam(model.parameters(), lr=1e-2, betas=(0.7, 0.99))
     criterion = nn.CrossEntropyLoss()
 
     for epoch in range(1, config.epoch + 1):
@@ -55,32 +56,60 @@ def train(config,model,train_iter, valid_iter,test_iter):
         torch.save(model.state_dict(), config.save_path)
 
 def test(config, model, TEXT, test_iter):
-
+    print("save_path: ", config.save_path)
+    model.load_state_dict(torch.load(config.save_path))
+    #print(model)
     b = next(iter(test_iter))
-    print("输入: ", b.text[0])
+    print(b.text.shape)
+    #print("输入: ", b.text[0])
     #print("输入的句子: ", word_ids_to_sentence(b.text[0],TEXT.vocab))
     #print("", word_sentence_to_ids(b.text[0],TEXT.vocab))
 
 
+    print("单条数据: ",b.text[:,1].shape)
 
     inputs_word = word_ids_to_sentence(b.text.cuda().data, TEXT.vocab)
-    print(inputs_word)
-    print(len(inputs_word))
+    #print(inputs_word)
+    #print(len(inputs_word))
 
     arrs = model(b.text.cuda()).cuda().data.cpu().numpy()
     print(arrs.shape)
     preds = word_ids_to_sentence(np.argmax(arrs, axis=2), TEXT.vocab)
+    return preds
+    #print(preds)
+
+def test_one_sentence(config, model , TEXT,test_iter):
+    print("save_path: ", config.save_path)
+    model.load_state_dict(torch.load(config.save_path))
+    #print(model)
+    b = next(iter(test_iter))
+    print(b.text.shape)
+    print("单条数据: ",b.text[:,1].shape)
+    print("单条数据: ",b.text[:,1].view(-1,1).shape)
+    inputs_word = word_ids_to_sentence(b.text[:,1].view(-1,1).cuda().data, TEXT.vocab)
+    print("inputs_word: ", inputs_word)
+    arrs = model(b.text[:,1].view(-1,1).cuda()).cuda().data.cpu().numpy()
+    preds = word_ids_to_sentence(np.argmax(arrs, axis=2), TEXT.vocab)
+    print("preds----------", preds)
 
-    print(preds)
 
 
 def test_sentence(config, model, TEXT, sentence):
+    model.load_state_dict(torch.load(config.save_path))
+    #print(model)
     inputs = torch.Tensor([TEXT.vocab.stoi[one] for one in sentence]).long().to(config.device)
+    print("inputs: ", inputs)
     inputs = inputs.view(-1, 1)
-    # print(inputs.shape)
+    #print("inputs: ", inputs)
+    print("inputs shape: ", inputs.shape)
+    #print(inputs)
     arrs = model(inputs)
     print("arrs shape: ",arrs.shape)
+    #print(arrs)
+    x=np.sum(np.array(arrs.detach().cpu()),axis=2)
     preds = word_ids_to_sentence(np.argmax(arrs.detach().cpu(), axis=2), TEXT.vocab)
+    #print(x)
+    print("preds: ",preds)
 
     return preds
 

__pycache__/utils.cpython-36.pyc

@@ -68,7 +68,8 @@ def __init__(self,config,TEXT):
 
         self.encoder=nn.Embedding(embeddings.shape[0],config.embedding_dim)
         self.decoder=nn.Linear(config.embedding_dim,embeddings.shape[0])
-        self.init_weights()
+        if config.mode!="test":
+            self.init_weights()
         config.n_tokens=embeddings.shape[0]
 
 
@@ -92,11 +93,16 @@ def forward(self,inputs,has_mask=True):
                 self.src_mask=mask
         else:
             self.src_mask=None
+        #print("1: ", inputs)
         inputs=self.encoder(inputs)*math.sqrt(self.embedding_dim)
         inputs=self.pos_encoder(inputs)
+        #print("2: ", inputs)
         output=self.transformer_encoder(inputs, self.src_mask)
+        #print("3: output: ", output)
         #print("output shape11: ", output.shape)
         output=self.decoder(output)
+        #print("4:", output)
         #print("output shape22: ", output.shape)
-        return F.log_softmax(output, dim=1)
+        #print(F.log_softmax(output, dim=2))
+        return F.log_softmax(output, dim=2)