RuntimeError: Expected tensor for argument #1 'indices' to have scalar type Long; but got torch.cuda.IntTensor instead (while checking arguments for embedding)

[Lydiasc34]

Hi, thank you for sharing the rcv data source and I managed to request data and preprocessed it. I used my preprocessed rcv text and its id as X_train/X_test, train_id/text_id in readData_rcv1. I chose TextCNN as base model and no pkl file exists so I started from the training.

And when processing data in load_data_rcv1, I noticed X_train = [[[word_index[word] for word in sent] for sent in doc] for doc in X_train]. So the type for X_train is list. Then the script works fine until it comes to the train_taxo() and below is the error I got.

(For the mode in config, I set it as hilap. And I use gpu to run.)

0it [00:00, ?it/s]
Traceback (most recent call last):
  File "C:/Users/HiLAP/main.py", line 816, in <module>
    train_taxo()
  File "C:/HiLAP/main.py", line 277, in train_taxo
    _, flat_probs = forward_step_sl(tokens, doc_ids, flat_probs_only=(args.global_ratio == 1))
  File "C:/HiLAP/main.py", line 134, in forward_step_sl
    probs = policy.base_model(tokens, True)
  File "C:\Local\Continuum\anaconda3\envs\HiLAP\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "C:\HiLAP\TextCNN.py", line 36, in forward
    x = self.embed(x)  # (N, W, D)
  File "C:\Local\Continuum\anaconda3\envs\HiLAP\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "C:\Local\Continuum\anaconda3\envs\HiLAP\lib\site-packages\torch\nn\modules\sparse.py", line 126, in forward
    self.norm_type, self.scale_grad_by_freq, self.sparse)
  File "C:\Local\Continuum\anaconda3\envs\HiLAP\lib\site-packages\torch\nn\functional.py", line 1852, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: Expected tensor for argument #1 'indices' to have scalar type Long; but got torch.cuda.IntTensor instead (while checking arguments for embedding)

It seems like the error is about input type. So I was trying to convert the tokens input to LongTensor but it keeps getting me errors. So I would like to show you the very first error I got to see if it was indeed the issue about tersor type. I've been researching and trying solutions but they did not work.

Looking forward to your reply. Thanks!

[Lydiasc34]

Also, even if I converted tokens to LongTensor ( and also some other variables such as mini_batch, cur_class_batch and next_classes_batch in the generate_logits function from model.py, that needs LongTensor type otherwise same error occurs), I got another error that:

0it [00:01, ?it/s]
Traceback (most recent call last):
File "C:/HiLAP/main.py", line 822, in
train_taxo()
File "C:/HiLAP/main.py", line 331, in train_taxo
finish_episode(policy, update=True)
File "C:/HiLAP/main.py", line 89, in finish_episode
entropy = torch.cat(policy.entropy_l).mean()
RuntimeError: zero-dimensional tensor (at position 0) cannot be concatenated

Have you ever had similar issues? Any thoughts would be appreciated!

[morningmoni]

I searched the first error, and the web suggested there might be some difference when running on Win and Linux. Sry I never tried on Win but I think the code should be runnable on Linux as ppl in other issues have reproduced the results.

[Perfect0530]

Hello,can you share your file"doc.txt"?

[morningmoni]

Hello,can you share your file"doc.txt"?

Hi, can you refer to https://github.com/morningmoni/HiLAP#data?