CoDi: Co-evolving Contrastive Diffusion Models for Mixed-type Tabular Synthesis

This code is the official implementation of "CoDi: Co-evolving Contrastive Diffusion Models for Mixed-type Tabular Synthesis". (https://arxiv.org/abs/2304.12654)

Requirements

Run the following to install requirements:

conda env create --file environment.yaml

Usage

• Train and evaluate CoDi through main.py:

main.py:
  --data: tabular dataset
  --eval : train or eval
  --logdir: Working directory

Training

• You can train our CoDi from scratch by run:

python main.py --data heart --logdir CoDi_exp

Evaluation

• By run the following script, you can reproduce our experimental result: binary classification result of CoDi on Heart in Table 10.

python main.py --data heart --eval True --logdir exp_heart