Fix an error on Windows about 32, 64-bit integer

[kkppll-ss]

I run translate.py on my Windows machine (with Windows 10 64-bit, Anaconda and PyTorch from this Anaconda package) and run into the following error:

Traceback (most recent call last):
  File "C:/Users/yao_z/Documents/attention-is-all-you-need-pytorch/translate.py", line 66, in <module>
    main()
  File "C:/Users/yao_z/Documents/attention-is-all-you-need-pytorch/translate.py", line 58, in main
    all_hyp, all_scores = translator.translate_batch(batch)
  File "C:\Users\yao_z\Documents\attention-is-all-you-need-pytorch\transformer\Translator.py", line 107, in translate_batch
    dec_partial_seq, dec_partial_pos, src_seq, enc_output)
  File "C:\Users\yao_z\Miniconda3\envs\pytorch\lib\site-packages\torch\nn\modules\module.py", line 357, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\yao_z\Documents\attention-is-all-you-need-pytorch\transformer\Models.py", line 111, in forward
    dec_input = self.tgt_word_emb(tgt_seq)
  File "C:\Users\yao_z\Miniconda3\envs\pytorch\lib\site-packages\torch\nn\modules\module.py", line 357, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\yao_z\Miniconda3\envs\pytorch\lib\site-packages\torch\nn\modules\sparse.py", line 103, in forward
    self.scale_grad_by_freq, self.sparse
  File "C:\Users\yao_z\Miniconda3\envs\pytorch\lib\site-packages\torch\nn\_functions\thnn\sparse.py", line 59, in forward
    output = torch.index_select(weight, 0, indices.view(-1))
TypeError: torch.index_select received an invalid combination of arguments - got (torch.cuda.FloatTensor, int, !torch.cuda.IntTensor!), but expected (torch.cuda.FloatTensor source, int dim, torch.cuda.LongTensor index)

This is because in the following code snippet from Beam.py, dec_seq is returned as a numpy ndarray with dtype=np.int32, not np.int64.

def get_tentative_hypothesis(self):
        "Get the decoded sequence for the current timestep."

        if len(self.next_ys) == 1:
            dec_seq = self.next_ys[0].unsqueeze(1)
        else:
            _, keys = self.sort_scores()
            hyps = [self.get_hypothesis(k) for k in keys]
            hyps = [[Constants.BOS] + h for h in hyps]
            dec_seq = torch.from_numpy(np.array(hyps))

        return dec_seq

On Windows, even 64 bit Windows and 64 bit Anaconda, numpy uses 32-bit integer as the default integral type, as is illustrated in the following picture and you can see more from here:

As a result, in the following code snippet from Translator.py, dec_partial_seq is an IntTensor instead of a LongTensor, which leads to the error I mentioned previously.

 # -- Decoding -- #
            dec_output, *_ = self.model.decoder(
dec_partial_seq, dec_partial_pos, src_seq, enc_output)

Therefore, I add dtype=np.int64 argument, the error disappears and it now runs smoothly on Windows.