embedding_not_training

Author: wabywang(王本友)

models/CNNBasic.py

@@ -14,7 +14,7 @@ def __init__(self, opt ):
 
         self.encoder = nn.Embedding(opt.vocab_size,opt.embedding_dim)
         if opt.__dict__.get("embeddings",None) is not None:
-            self.encoder.weight=nn.Parameter(opt.embeddings)
+            self.encoder.weight=nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 
         self.content_conv = nn.Sequential(
             nn.Conv1d(in_channels = opt.embedding_dim,

models/CNNMultiLayer.py

@@ -12,7 +12,7 @@ def __init__(self, opt):
         self.embed = nn.Embedding(opt.vocab_size + 1, opt.embedding_dim)
 
         if opt.__dict__.get("embeddings",None) is not None:
-            self.embed.weight=nn.Parameter(opt.embeddings)
+            self.embed.weight=nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 
         self.conv1 = nn.Sequential(
             nn.Conv1d(opt.max_seq_len, 256, kernel_size=7, stride=1),

models/CNNText.py

@@ -14,7 +14,7 @@ def __init__(self, opt ):
 
         self.encoder = nn.Embedding(opt.vocab_size,opt.embedding_dim)
         if opt.__dict__.get("embeddings",None) is not None:
-            self.encoder.weight=nn.Parameter(opt.embeddings)
+            self.encoder.weight=nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 
 
         self.content_conv = nn.Sequential(

models/Capsule.py

@@ -86,7 +86,7 @@ def __init__(self,opt):
         self.kernel_size = 3
         self.kernel_size_primary=3
         if opt.__dict__.get("embeddings",None) is not None:
-            self.embed.weight=nn.Parameter(opt.embeddings)
+            self.embed.weight=nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 
         self.primary_capsules = CapsuleLayer(num_capsules=8, num_route_nodes=-1, in_channels=256, out_channels=32)
         self.digit_capsules = CapsuleLayer(num_capsules=opt.label_size, num_route_nodes=int(32 * opt.max_seq_len/2), in_channels=8,

models/FastText.py

@@ -14,7 +14,7 @@ def __init__(self, opt ):
         self.encoder = nn.Embedding(opt.vocab_size,opt.embedding_dim)
         if opt.__dict__.get("embeddings",None) is not None:
             print('load embedding')
-            self.encoder.weight=nn.Parameter(opt.embeddings)
+            self.encoder.weight=nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 
 
         self.content_fc = nn.Sequential(

models/LSTM.py

@@ -16,7 +16,7 @@ def __init__(self,opt):
         self.use_gpu = torch.cuda.is_available()
 
         self.word_embeddings = nn.Embedding(opt.vocab_size, opt.embedding_dim)
-        self.word_embeddings.weight = nn.Parameter(opt.embeddings)
+        self.word_embeddings.weight = nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 #        self.word_embeddings.weight.data.copy_(torch.from_numpy(opt.embeddings))
         self.lstm = nn.LSTM(opt.embedding_dim, opt.hidden_dim)
         self.hidden2label = nn.Linear(opt.hidden_dim, opt.label_size)

models/LSTMBI.py

@@ -16,7 +16,7 @@ def __init__(self,opt):
         self.use_gpu = torch.cuda.is_available()
 
         self.word_embeddings = nn.Embedding(opt.vocab_size, opt.embedding_dim)
-        self.word_embeddings.weight = nn.Parameter(opt.embeddings)
+        self.word_embeddings.weight = nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 #        self.word_embeddings.weight.data.copy_(torch.from_numpy(opt.embeddings))
 
         self.lstm_layers = opt.lstm_layers

models/LSTMwithAttention.py

@@ -15,7 +15,7 @@ def __init__(self,opt):
         self.use_gpu = torch.cuda.is_available()
 
         self.word_embeddings = nn.Embedding(opt.vocab_size, opt.embedding_dim)
-        self.word_embeddings.weight = nn.Parameter(opt.embeddings)
+        self.word_embeddings.weight = nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 #        self.word_embeddings.weight.data.copy_(torch.from_numpy(opt.embeddings))
 
         self.num_layers = opt.lstm_layers

models/RCNN.py

@@ -18,7 +18,7 @@ def __init__(self,opt):
         self.use_gpu = torch.cuda.is_available()
 
         self.word_embeddings = nn.Embedding(opt.vocab_size, opt.embedding_dim)
-        self.word_embeddings.weight = nn.Parameter(opt.embeddings)
+        self.word_embeddings.weight = nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 #        self.word_embeddings.weight.data.copy_(torch.from_numpy(opt.embeddings))
 
         self.num_layers = 1

models/RNN_CNN.py

@@ -14,7 +14,7 @@ def __init__(self,opt):
         self.use_gpu = torch.cuda.is_available()
 
         self.word_embeddings = nn.Embedding(opt.vocab_size, opt.embedding_dim)
-        self.word_embeddings.weight = nn.Parameter(opt.embeddings)
+        self.word_embeddings.weight = nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 #        self.word_embeddings.weight.data.copy_(torch.from_numpy(opt.embeddings))
         self.lstm = nn.LSTM(opt.embedding_dim, opt.hidden_dim)
         ###self.hidden2label = nn.Linear(opt.hidden_dim, opt.label_size)

models/SelfAttention.py

@@ -19,7 +19,7 @@ def __init__(self,opt):
         self.use_gpu = torch.cuda.is_available()
 
         self.word_embeddings = nn.Embedding(opt.vocab_size, opt.embedding_dim)
-        self.word_embeddings.weight = nn.Parameter(opt.embeddings)
+        self.word_embeddings.weight = nn.Parameter(opt.embeddings,requires_grad=opt.embedding_training)
 #        self.word_embeddings.weight.data.copy_(torch.from_numpy(opt.embeddings))
 
         self.num_layers = 1

opts.py

@@ -30,6 +30,8 @@ def parse_opt():
                     help='max_epoch')
     parser.add_argument('--embedding_file', type=str, default="glove.6b.300",
                     help='glove or w2v')
+    parser.add_argument('--embedding_training', type=str, default="false",
+                    help='embedding_training')
     #kim CNN
     parser.add_argument('--kernel_sizes', type=str, default="1,2,3,5",
                     help='kernel_sizes')
@@ -50,6 +52,9 @@ def parse_opt():
 
 
 
+    
+    
+    
 #
     args = parser.parse_args()
 
@@ -70,6 +75,11 @@ def parse_opt():
         args.debug = True
     else:
         args.debug = False
+    
+    if args.embedding_training.lower() =="true":
+        args.embedding_training = True
+    else:
+        args.embedding_training = False
     if os.path.exists("proxy.config"):
         with open("proxy.config") as f: