This repository is the implementation for paper Adversarial Text Generation via Feature-Mover’s Distance

Python 2.7, Tensorflow 1.8.0

• Run: python autoencoder.py for MLE pre-train
• Run: python text_GAN.py for adversarial training
• Options: options can be made by changing option class.

model.py: sinkhorn divergence edition

model2.py: IPOT edition

MSCOCO dataset and WMT news dataset can be downloaded from link HERE: Download link

Use convert_new.py to convert the indexed files to sentences. Then use selfbleu.py and testbleu.py to evaluate the results.

Note that it is just a rough edition, you have to change the file names manually in the code, we will update this ASAP.

1. Clean the code, make it easy to understand.
2. Don't need to manually change the file names in the code.
3. etc.

Please cite our paper if it helps with your research

@inproceedings{chen2018adversarial,
  title={Adversarial Text Generation via Feature-Mover's Distance},
  author={Chen, Liqun and Dai, Shuyang and Tao, Chenyang and Shen, Dinghan and Gan, Zhe and Zhang, Haichao and Zhang, Yizhe and Carin, Lawrence},
  Booktitle={NIPS},
  year={2018}
}

For any question or suggestions, feel free to contact my email.