/SimpDM

[CIKM'24] Self-Supervision Improves Diffusion Models for Tabular Data Imputation

Primary LanguagePythonMIT LicenseMIT

Self-Supervision Improves Diffusion Models for Tabular Data Imputation

This is the source code of CIKM'24 paper "Self-Supervision Improves Diffusion Models for Tabular Data Imputation" (SimpDM).

The proposed framework

USAGE

Step 1. Create a new conda environment with Python 3.10:

conda create -n simpdm python=3.10

Step 2. Initialize the environment by running:

bash init_env.sh

Step 3. Run the corresponding script in “run.sh”. For example, for Yacht dataset we can run the first line:

python3 main.py --dataset yacht --epochs 30000 --lr 0.0001 --num_layers 5 --hidden_units 256 --gammas 1.0_0.8_0.001 --ssl_loss_weight 10

CITE

If you compare with, build on, or use aspects of SimpDM framework, please cite the following:

@article{liu2024self,
  title={Self-supervision improves diffusion models for tabular data imputation},
  author={Liu, Yixin and Ajanthan, Thalaiyasingam and Husain, Hisham and Nguyen, Vu},
  journal={arXiv preprint arXiv:2407.18013},
  year={2024}
}