A PyTorch attempt at reimplementing

• MaskGAN: Better Text Generation via Filling in the _______ , William Fedus, Ian Goodfellow, Andrew M. Dai [paper]

Pytorch1.1 Python3

Some of you may encounter "no attribute 'IterableDataset", please check your pytorch version and use pip install, not conda.

Some devices may enconter : module:command not find

sudo apt-get install environment-modules

python3 -m pip install tqdm

I used google/SentencePiece to bring down the vocabulary to make training easier. The trained models are available inside this repository. Install the python bindings through pip so the code can use it.

python3 -m pip install sentencepiece

This code is build using the basic blocks provided by pytorch/fairseq. Please follow instructions there to install fairseq as a library.

python3 -m pip install git+https://github.com/pytorch/fairseq

mkdir datasets 
cd datasets
IMDB_DATASET='http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz'
wget $IMDB_DATASET -O aclImdb_v1.tar.gz
tar xvzf aclImdb_v1.tar.gz

Launch a visdom instance for logging. If you enconter some HTTP errors, perhaps it's visdom's problem. Try to restart it or change the port :)

python3 -m pip install visdom # Install if not present.
python3 -m visdom.server &

Run the training script.

python3 -m mgan.main \
  --path datasets/aclImdb/train/ \
  --spm_path datasets/aclImdb/train/imdb