Composite Evolutionary GAN for Natural Language Generation with Temper Control

@INPROCEEDINGS{9674611,
author={Sun, Fanglei and Tao, Qian and Hu, Jianqiao and Liu, Jieqiong},
booktitle={2021 7th International Conference on Computer and Communications (ICCC)},
title={Composite Evolutionary GAN for Natural Language Generation with Temper Control},
year={2021},
volume={},
number={},
pages={1710-1714},
doi={10.1109/ICCC54389.2021.9674611}}

Requirements

PyTorch >= 1.1.0
Python 3.6
Numpy 1.14.5
CUDA 7.5+ (For GPU)
nltk 3.4
tqdm 4.32.1
KenLM (https://github.com/kpu/kenlm)

To install, run pip install -r requirements.txt. In case of CUDA problems, consult the official PyTorch Get Started guide.

KenLM Installation

Download stable release and unzip: http://kheafield.com/code/kenlm.tar.gz
Need Boost >= 1.42.0 and bjam
- Ubuntu: sudo apt-get install libboost-all-dev
- Mac: brew install boost; brew install bjam

Run within kenlm directory:

mkdir -p build
cd build
cmake ..
make -j 4

pip install https://github.com/kpu/kenlm/archive/master.zip
For more information on KenLM see: https://github.com/kpu/kenlm and http://kheafield.com/code/kenlm/

Implemented Models and Original Papers

General Text Generation

SeqGAN - SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient
LeakGAN - Long Text Generation via Adversarial Training with Leaked Information
MaliGAN - Maximum-Likelihood Augmented Discrete Generative Adversarial Networks
JSDGAN - Adversarial Discrete Sequence Generation without Explicit Neural Networks as Discriminators
RelGAN - RelGAN: Relational Generative Adversarial Networks for Text Generation
DPGAN - DP-GAN: Diversity-Promoting Generative Adversarial Network for Generating Informative and Diversified Text
DGSAN - DGSAN: Discrete Generative Self-Adversarial Network
CoT - CoT: Cooperative Training for Generative Modeling of Discrete Data

Category Text Generation

SentiGAN - SentiGAN: Generating Sentimental Texts via Mixture Adversarial Networks
CatGAN (ours) - CatGAN: Category-aware Generative Adversarial Networks with Hierarchical Evolutionary Learning for Category Text Generation

Get Started

Get Started

git clone https://github.com/williamSYSU/TextGAN-PyTorch.git
cd TextGAN-PyTorch

For real data experiments, all datasets (Image COCO, EMNLP NEWs, Movie Review, Amazon Review) can be downloaded from here.
Run with a specific model

cd run
python3 run_[model_name].py 0 0	# The first 0 is job_id, the second 0 is gpu_id

# For example
python3 run_seqgan.py 0 0

Features

Instructor

For each model, the entire runing process is defined in instructor/oracle_data/seqgan_instructor.py. (Take SeqGAN in Synthetic data experiment for example). Some basic functions like init_model()and optimize() are defined in the base class BasicInstructor in instructor.py. If you want to add a new GAN-based text generation model, please create a new instructor under instructor/oracle_data and define the training process for the model.
Visualization

Use utils/visualization.py to visualize the log file, including model loss and metrics scores. Custom your log files in log_file_list, no more than len(color_list). The log filename should exclude .txt.
Logging

The TextGAN-PyTorch use the logging module in Python to record the running process, like generator's loss and metric scores. For the convenience of visualization, there would be two same log file saved in log/log_****_****.txt and save/**/log.txt respectively. Furthermore, The code would automatically save the state dict of models and a batch-size of generator's samples in ./save/**/models and ./save/**/samples per log step, where ** depends on your hyper-parameters.
Running Signal

You can easily control the training process with the class Signal (please refer to utils/helpers.py) based on dictionary file run_signal.txt.

For using the Signal, just edit the local file run_signal.txt and set pre_sig to Fasle for example, the program will stop pre-training process and step into next training phase. It is convenient to early stop the training if you think the current training is enough.
Automatiaclly select GPU

In config.py, the program would automatically select a GPU device with the least GPU-Util in nvidia-smi. This feature is enabled by default. If you want to manually select a GPU device, please uncomment the --device args in run_[run_model].py and specify a GPU device with command.