JayoungKim408 / STaSy

Official code for "STaSy: Score-based Tabular data Synthesis", ICLR 2023

STaSy: Score-based Tabular Data Synthesis

This code is the official implementation of "STaSy: Score-based Tabular Data Synthesis"(https://arxiv.org/abs/2210.04018).

Link for Tutorial:

• Google Colab Notebook Link: Colab

Requirements

Run the following to install requirements:

conda env create --file requirements.yaml

Usage

• Train, fine-tune, and evaluate our STaSy through main.py:

main.py:
  --config: Training configuration.
  --mode: <train|fine_tune|eval>: Running mode: train or fine_tune or eval
  --workdir: Working directory

Training

• You can train our STaSy with SPL from scratch by run:

python main.py --config configs/shoppers.py --mode train --workdir stasy

• To fine-tune the model trained with SPL by run:

python main.py --config configs/shoppers.py --mode fine_tune --workdir stasy --config.optim.lr 2e-07

Evaluation

• You can download pretrained model from the anonymous link here.
• Downloaded checkpoint should be in 'stasy/checkpoints/'.
• By run the following script, you can reproduce our experimental result: binary classification result of STaSy on Shoppers in Table 11.

python main.py --config configs/shoppers.py --mode eval --workdir stasy