Figure 1: TabRep Architecture.
Figure 2: TabRep Training and Sampling Algorithms.
Our implementation is built upon TabSYN's repository. The following delineates the installation instructions.
Clone this repository and navigate to it in your terminal.
Create an environment using a preferred package manager. We provide installation instructions for Conda and UV. We recommend using UV to for faster installation. This environment can be used for everything apart from eval_quality:
Conda Installation:
conda create -n tabrep python=3.10
conda activate tabrep
conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install -r requirements.txt
pip install dgl -f https://data.dgl.ai/wheels/cu117/repo.html
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.0.1+cu117.html
UV Installation:
uv venv --python 3.10
source .venv/bin/activate
uv pip install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2+cu117 --extra-index-url https://download.pytorch.org/whl/cu117
uv pip install -r requirements.txt
uv pip install dgl -f https://data.dgl.ai/wheels/cu117/repo.html
uv pip install torch_geometric
uv pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.0.1+cu117.html
Any other missing dependencies can be installed using pip. Once all the dependencies are installed, the scripts should run accordingly.
For eval_quality, create the following environment:
Conda Installation:
conda create -n tabrep_quality python=3.10
conda activate tabrep_quality
python -m pip install "pip<24.1"
pip install synthcity
pip install icecream
pip install category_encoders
pip install tomli
pip install tomli_w
pip install zero
pip install scipy
pip install click==8.1.7
pip install scipy==1.12.0
pip install --upgrade pip
UV Installation:
uv venv tabrep_quality --python 3.10 --seed
source tabrep_quality/bin/activate
uv pip install "pip<24.1"
uv pip install synthcity
uv pip install icecream
uv pip install category_encoders
uv pip install tomli
uv pip install tomli_w
uv pip install zero
uv pip install scipy==1.12.0
uv pip install click==8.1.7
uv pip install --upgrade pip
Download baseline datasets and run them within the tabrep_main directory as follows:
python download_dataset.py
python process_dataset.py
Classification:
- Adult: https://archive.ics.uci.edu/dataset/2/adult
- Default of Credit Card Clients: https://archive.ics.uci.edu/dataset/350/default+of+credit+card+clients
- Stroke Prediction: https://www.kaggle.com/datasets/fedesoriano/stroke-prediction-dataset
- Shoppers Purchasing Intention: https://archive.ics.uci.edu/dataset/468/online+shoppers+purchasing+intention+dataset
Multiclass-Classification:
- Diabetes Readmissions: https://archive.ics.uci.edu/dataset/296/diabetes+130-us+hospitals+for+years+1999-2008
Regression:
- Beijing PM2.5: https://archive.ics.uci.edu/dataset/381/beijing+pm2+5+data
- News Popularity: https://archive.ics.uci.edu/dataset/332/online+news+popularity
As Kaggle requires user log-on, the Stroke dataset should be downloaded manually.
To process Kaggle datasets:
cd data
mkdir [dataname]
# Move the downloaded Kaggle dataset .csv file into the newly-created folder
cd ..
python process_dataset.py [dataname]
We include the following baseline methods and datasets:
Baselines: codi, ctgan, goggle, great, smote, stasy, tabddpm, tabsyn
Datasets: adult, default, stroke, shoppers, diabetes, beijing, news
Training:
python main.py --dataname [NAME_OF_DATASET] --method [NAME_OF_BASELINE_METHODS] --mode train
python main.py --dataname adult --method tabrep_ddpm --mode train
python main.py --dataname adult --method tabrep_flow --mode train
Sampling:
python main.py --dataname [NAME_OF_DATASET] --method [NAME_OF_BASELINE_METHODS] --mode sample
python main.py --dataname adult --method tabrep_ddpm --mode sample
python main.py --dataname adult --method tabrep_flow --mode sample
Evaluation:
We include Machine Learning Efficiency (AUC, RMSE, F1 Scores), Density Estimation (Pair-Wise Column Correlation/Column Density Estimation), Quality (
For eval_privacy.py, an extra step is required to setup the package. Within the syntheval pip installation in the original tabrep environment (i.e. python3.10/site-packages/syntheval/presets), create a new mia.json file and write the following:
{
"mia" : {"num_eval_iter": 5}
}
Then evaluate as follows:
python eval/eval_mle.py --dataname [NAME_OF_DATASET] --model [NAME_OF_BASELINE_METHODS] --path [PATH_TO_SYNTHETIC_DATA]
python eval/eval_density.py --dataname [NAME_OF_DATASET] --model [METHOD_NAME] --path [PATH_TO_SYNTHETIC_DATA]
python eval/eval_quality.py --dataname [NAME_OF_DATASET] --model [METHOD_NAME] --path [PATH_TO_SYNTHETIC_DATA]
python eval/eval_detection.py --dataname [NAME_OF_DATASET] --model [METHOD_NAME] --path [PATH_TO_SYNTHETIC_DATA]
python eval/eval_privacy.py --dataname [NAME_OF_DATASET] --model [METHOD_NAME] --path [PATH_TO_SYNTHETIC_DATA]
python eval/eval_mle.py --dataname beijing --model tabrep_ddpm --path synthetic/beijing/tabrep_ddpm.csv
python eval/eval_mle.py --dataname beijing --model tabrep_flow --path synthetic/beijing/tabrep_flow.csv
Please consider citing our paper and giving it a 🌟 if you find it helpful. Thank you 😀!
@article{si2025tabrep,
title={TabRep: Training Tabular Diffusion Models with a Simple and Effective Continuous Representation},
author={Si, Jacob and Ou, Zijing and Qu, Mike and Xiang, Zhengrui and Li, Yingzhen},
journal={arXiv preprint arXiv:2504.04798},
year={2025}
}