zhengliz/natural-adversary

Generating Natural Adversarial Examples, ICLR 2018

natural-adversary's Issues

Is kenlm necessary

I found kenlm in the text code, but kenlm training will not be excuted with default parameters setting. Is kenlm training necessary for the generating fluent text?

Training encoder from deliminator loss?

Hi, thank you for sharing the code. I read your paper and I am trying to train and modify.
I have a question about how you optimize the encoder.
You are optimizing the encoder and the critic from the same loss at train_gan_d, but is this collect?
In my understanding, the encoder's loss function and the critic loss function are different.

There is no requirements.txt or .yml file for recreating the environment for this code to run in. I'm guessing this is python 2 since I'm getting issues with python 3. I also don't know what pytorch release was used.

Text: missing data labels

Hello there,

First, thanks for sharing your code!

I'm trying to run train_baseline for text but I'm getting this error:

Traceback (most recent call last):
  File "train_baseline.py", line 66, in <module>
    corpus_train = SNLIDataset(train=True, vocab_size=11004, lvt=False, path=args.data_path)
  File "/home/aosman/02_fields/adversarial/natural-adversary/text/utils.py", line 276, in __init__
    self.train_data = self.tokenize(self.train_path)
  File "/home/aosman/02_fields/adversarial/natural-adversary/text/utils.py", line 336, in tokenize
    label = self.labels[tokens[0]]
KeyError: 'A person on a horse jumps over a broken down airplane .'

it seems that the labels are missing from the text files (train.txt and test.txt). Would you be kind to provide them or let me know where to get them?

Thanks!

generate.py throwing issue

Hi zhengliz,

Thanks for sharing code.

just followed README.
i was trying see demo results. but couldn't get it as it's throwing issue.

$ python generate.py --load_path ./output/1535710044
{'ninterpolations': 5, 'temp': 1, 'load_path': './output/1535710044', 'sample': False, 'seed': 1111, 'steps': 5, 'noprint': False, 'outf': './generated.txt', 'ngenerations': 10}
Loading models from./output/1535710044/models
Traceback (most recent call last):
  File "generate.py", line 135, in <module>
    main(args)
  File "generate.py", line 74, in main
    maxlen=model_args['maxlen'])
  File "/natural-adversary/text/models.py", line 654, in generate
    sample=sample)
  File "/natural-adversary/text/models.py", line 303, in generate
    embedding = self.embedding_decoder(self.start_symbols)
  File "/venv/local/lib/python2.7/site-packages/torch/nn/modules/module.py", line 357, in __call__
    result = self.forward(*input, **kwargs)
  File "/venv/local/lib/python2.7/site-packages/torch/nn/modules/sparse.py", line 103, in forward
    self.scale_grad_by_freq, self.sparse
  File "/venv/local/lib/python2.7/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.LongTensor), but expected (torch.cuda.FloatTensor source, int dim, torch.cuda.LongTensor index)

could you please tell me why is this issue coming ?

thanks in advance.