dsksd/rnn-for-joint-nlu

Pytorch implementation of "Attention-Based Recurrent Neural Network Models for Joint Intent Detection and Slot Filling" (https://arxiv.org/abs/1609.01454)

rnn-for-joint-nlu's Issues

There aren't models documents that contains the jointnlu-encoder.pkl and jointnlu-decoder.pkl model

FileNotFoundError: [Errno 2] No such file or directory: 'models/jointnlu-encoder.pkl'****
FileNotFoundError: [Errno 2] No such file or directory: 'models/jointnlu-decoder.pkl'****

There aren't models documents that contains the jointnlu-encoder.pkl and jointnlu-decoder.pkl model, could you give me a solution?

code question in train.py

Hi,thank you for implementing the code for that paper. But I encountered a problem while

 tag_score, intent_score = decoder(start_decode,hidden_c,output,x_mask)

> Traceback (most recent call last):
  File "E:/NER/RNN-for-Joint-NLU-master/train.py", line 102, in <module>
    train(config)
  File "E:/NER/RNN-for-Joint-NLU-master/train.py", line 54, in train
    tag_score, intent_score = decoder(start_decode,hidden_c,output,x_mask)
  File "D:\anaconda3\lib\site-packages\torch\nn\modules\module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "E:\NER\RNN-for-Joint-NLU-master\model.py", line 120, in forward
    _, hidden = self.lstm(torch.cat((embedded,context,aligned),2), hidden) # input, context, aligned encoder hidden, hidden
  File "D:\anaconda3\lib\site-packages\torch\nn\modules\module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "D:\anaconda3\lib\site-packages\torch\nn\modules\rnn.py", line 175, in forward
    self.check_forward_args(input, hx, batch_sizes)
  File "D:\anaconda3\lib\site-packages\torch\nn\modules\rnn.py", line 152, in check_forward_args
    'Expected hidden[0] size {}, got {}')
  File "D:\anaconda3\lib\site-packages\torch\nn\modules\rnn.py", line 148, in check_hidden_size
    raise RuntimeError(msg.format(expected_hidden_size, tuple(hx.size())))
RuntimeError: Expected hidden[0] size (1, 16, 128), got (2, 16, 128)

Could not set num_layers > 1 for LSTM

since in the code we suppose the hidden layer's shape is [1, B, D], if we set num_layers > 1 for LSTM whose hidden layer's shape will be [>1, B, D] so that some continued operation could not be executed successfully.

Intent score

Thank you for open sourcing your code. I couldn't reach the intent perfomance mentioned in the paper. Can you tell what is a performance that you getting?

Thanks in advance.

clone

hi, in the class Decoder, i want to know when u compute the intent hidden, why use copy()? why not directly use the hidden[0]?