New pytorch + embedding fix

Nets.py

@@ -32,7 +32,7 @@ def forward(self, packed_batch):
         all_att = self._masked_softmax(attend,self._list_to_bytemask(list(len_s))).transpose(0,1) # attW,sent 
         attended = all_att.unsqueeze(2).expand_as(enc_sents) * enc_sents
 
-        return attended.sum(0).squeeze(0)
+        return attended.sum(0,True).squeeze(0)
 
     def _list_to_bytemask(self,l):
         mask = self._buffers['mask'].resize_(len(l),l[0]).fill_(1)
@@ -45,7 +45,7 @@ def _list_to_bytemask(self,l):
 
     def _masked_softmax(self,mat,mask):
         exp = torch.exp(mat) * Variable(mask)
-        sum_exp = exp.sum(1)+0.0001
+        sum_exp = exp.sum(1,True)+0.0001
 
         return exp/sum_exp.expand_as(exp)
 

main.py

@@ -154,7 +154,7 @@ def test(epoch,net,dataset,cuda):
 def accuracy(out,truth):
     def sm(mat):
         exp = torch.exp(mat)
-        sum_exp = exp.sum(1)+0.0001
+        sum_exp = exp.sum(1,True)+0.0001
         return exp/sum_exp.expand_as(exp)
 
     _,max_i = torch.max(sm(out),1)
@@ -206,7 +206,16 @@ def main(args):
     print(25*"-" + "\nBuilding word vectors: \n"+"-"*25)
 
     vectorizer = Vectorizer(max_word_len=args.max_words,max_sent_len=args.max_sents)
-    vectorizer.build_dict(train_set.field_iter(0),args.max_feat)
+
+    if args.emb:
+        tensor,dic = load_embeddings(args.emb)
+        net = HierarchicalDoc(ntoken=len(dic),num_class=num_class)
+        net.set_emb_tensor(torch.FloatTensor(tensor))
+        vectorizer.word_dict = dic
+    else:
+        vectorizer.build_dict(train_set.field_iter(0),args.max_feat)
+        net = HierarchicalDoc(ntoken=len(vectorizer.word_dict),num_class=num_class)
+
 
     tuple_batch = tuple_batcher_builder(vectorizer,train=True)
     tuple_batch_test = tuple_batcher_builder(vectorizer,train=False)
@@ -216,20 +225,19 @@ def main(args):
     sampler = None
     if args.balance:
         sampler =  BucketSampler(train_set)
+        sampler_t = BucketSampler(test_set)
 
 
-    dataloader = DataLoader(train_set, batch_size=args.b_size, shuffle=True, sampler=sampler, num_workers=1, collate_fn=tuple_batch)#, collate_fn=<function default_collate>, pin_memory=False)
-    dataloader_test = DataLoader(test_set, batch_size=args.b_size, shuffle=True,  num_workers=1, collate_fn=tuple_batch_test)#, collate_fn=<function default_collate>, pin_memory=False)
+    dataloader = DataLoader(train_set, batch_size=args.b_size, shuffle=False, sampler=sampler, num_workers=1, collate_fn=tuple_batch)#, collate_fn=<function default_collate>, pin_memory=False)
+    dataloader_test = DataLoader(test_set, batch_size=args.b_size, shuffle=False,sampler=sampler_t,  num_workers=1, collate_fn=tuple_batch_test)#, collate_fn=<function default_collate>, pin_memory=False)
 
 
     criterion = torch.nn.CrossEntropyLoss()
 
 
 
-    net = HierarchicalDoc(ntoken=len(vectorizer.word_dict),num_class=num_class)
 
-    if args.emb:
-        net.set_emb_tensor(torch.FloatTensor(tensor))
+
 
     if args.cuda:
         net.cuda()