anubhavgupta3377 / text-classification-models-pytorch Goto Github PK

use Model_Transformer
when I run train.py,I got as follow:
TypeError: forward() missing 1 required positional argument: 'mask'

Can you give me the reason?

Hi, thanks for your code! However, I think there might be a bug on NLLLoss: I think the input of nn.NLLloss() is after logsoftmax according to this manual: https://pytorch.org/docs/stable/generated/torch.nn.NLLLoss.html, however, here only softmax is provided in model.py therefore, the loss is always negative. I only check the Transformer model.

Kind regards,
John

Hi,
What are the files named "ag_news.train" and "ag_news.test"?

Hi, when i run the train.py in the folder "textCNN model",there is an error that I cant find out what's wrong. Can you help me ? Thank you very much

Loaded 96000 training examples
Loaded 7600 test examples
Loaded 24000 validation examples
Epoch: 0
Traceback (most recent call last):
  File "/Users/y/Documents/Code/Text-Classification-Models-Pytorch-master/Model_TextCNN/train.py", line 43, in <module>
    train_loss,val_accuracy = model.run_epoch(dataset.train_iterator, dataset.val_iterator, i)
  File "/Users/y/Documents/Code/Text-Classification-Models-Pytorch-master/Model_TextCNN/model.py", line 85, in run_epoch
    y_pred = self.__call__(x)
  File "/anaconda3/python.app/Contents/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/Users/y/Documents/Code/Text-Classification-Models-Pytorch-master/Model_TextCNN/model.py", line 45, in forward
    embedded_sent = self.embeddings(x).permute(1,2,0)
  File "/anaconda3/python.app/Contents/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/anaconda3/python.app/Contents/lib/python3.6/site-packages/torch/nn/modules/sparse.py", line 118, in forward
    self.norm_type, self.scale_grad_by_freq, self.sparse)
  File "/anaconda3/python.app/Contents/lib/python3.6/site-packages/torch/nn/functional.py", line 1454, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: index out of range at /Users/administrator/nightlies/pytorch-1.0.0/wheel_build_dirs/conda_3.6/conda/conda-bld/pytorch_1544137972173/work/aten/src/TH/generic/THTensorEvenMoreMath.cpp:191

Process finished with exit code 1

hi, setting max_length to none, but short sentence would still be padded to the longest length among the batch, thus affects the training and prediction result
eg.
predict only one sentence, "0516酸菜鱼", index input to tensor([[241542],
[ 7789],
[ 192],
[ 260]])
but add "'Bird&Bird 香港鸿鹄律师事务所北京代表处鸿鹄知识产权代理(北京有限公司)'
index input to ":
whereas index 1 is "<pad>"
and output differs:

I don't know why , I just add the code in config.py
max_sen_len = None

can you give me some ideas,thanks

On lines 53 and 61 of Text-Classification-Models-Pytorch/Model_RCNN/model.py , function permute is used.

input_features = torch.cat([lstm_out,embedded_sent], 2).permute(1,0,2)
...
linear_output = linear_output.permute(0,2,1) # Reshaping fot max_pool

Could you please explain why it is necessary or useful to permute the dimensions of these tensors?

In fasttext model:
def forward(self, x):
embedded_sent = self.embeddings(x).permute(1,0,2)
h = self.fc1(embedded_sent.mean(1))
z = self.fc2(h)
return self.softmax(z)

Why is there no call to relu on h?

Hi, would you please share your point in "Only encoder part of Transformer model is used for classification",.
I apply this model in my dataset, the accuracy only 60% .I am thinking whether the number of layer is a cause?