This is the code for the project Causal-TGAN: Causally-Aware Tabular Data Generative Adversarial Networks
If you use this code for your research, please cite:
@inproceedings{
wen2022causaltgan,
title={Causal-{TGAN}: Modeling Tabular Data Using Causally-Aware {GAN}},
author={Bingyang Wen and Yupeng Cao and Fan Yang and Koduvayur Subbalakshmi and Rajarathnam Chandramouli},
booktitle={ICLR Workshop on Deep Generative Models for Highly Structured Data},
year={2022},
url={https://openreview.net/forum?id=BEhxCh4dvW5}
}
The project is built on python3.6 with the pytorch version of 1.9.0.
To run this project, please add a new python environment path to this project. For example if your cloned repository reside in /home/username/CausalTGAN
, then one way to do this is export PYTHONPATH="/home/username"
from command line or add it to your ~/.bashrc
.
-
An example dataset is illustrated in
./data/real_world/adult
. -
Instructions of adding the customized dataset is following:
- Step 1: Add your dataset name and name of column into
helper/constant.py
- Step 2: Identify your data type (Continuous, Discrete, Mix) and
- Step 1: Add your dataset name and name of column into
python train.py
. The training details such as dataset and epochs are set inside train.py
python sampling.py
. The details such as number of samples to generate and causal-TGAN path are set inside sampling.py