SeqGAN-Colaboratory

This repository is forked from https://github.com/MiyoshiYuto/SeqGAN. Thank you MiyoshiYuto for rewriting the SeqGan code to run in a Google Colaboratory compatible environment!

Requirements

• Tensorflow r2.2.0
• Python 3.6+
• CUDA 10.1 (For GPU)
• This is compatible with the default GPU environment in Google Colaboratory :) (2020/01/28)

Note: this code is based on the previous work by ofirnachum. Many thanks to ofirnachum.

Paper

SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient has been accepted at the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17).

Motivation

This fork is intended to be used to train a SeqGan model for free in Google Colaboratory on a custom dataset (eg. movie reviews). I hope that this makes the original art more accessible.

File descriptions

In addition to the original code, I have added the following files:

• prepare_nltk_data.ipynb - Download the nltk movie review dataset and save for cunsumption by sequence_gan.py (for SeqGan model training).
• prepare_custom_data.ipynb, prepare_custom_data.py - Code to prepare a custom text dataset for cunsumption by sequence_gan.py.
• seq_gan_train.ipynb - An example Colab notebook for SeqGan training. Use Colab's free GPU environment!
• seq_gan_run.ipynb - An example Colab notebook for generating new sequences using a saved SeqGan.

I have also removed the originat TARGET_LSTM oracle model and .pkl since this code is intended to be used on real text data. An example of movie review text data is svaed in ./data/. You can also create your own data using either prepare_nltk_data.ipynb, prepare_custom_data.ipynb or prepare_custom_data.py.

How to train a model

Simply upload the notebook seq_gan_train.ipynb to Colaboratory and run it!