PrivATE

Implementation of PrivATE

Requirements

numpy >= 1.24.4
pandas >= 2.0.3
scikit-learn >= 1.3.2
scipy >= 1.10.1
python >= 3.8

Introduction

Below is the directory structure of this project.

PrivATE/
├── dataset/                         # Datasets used in the project
│   ├── ACIC.csv     
│   ├── ACIC.json   
│   ├── IHDP.csv       
│   ├── IHDP.json 
│   ├── Lalonde.csv
│   ├── Lalonde.json
│   ├── Synth.csv            
│   ├── Synth.json    
├── main_label.py                    # Implementation for label-level privacy of PrivATE
├── main_sample.py                   # Implementation for sample-level privacy of PrivATE               
├── README.md                        # Main project documentation                                           
├── utils.py                         # Helper classes and functions supporting various operations                  

Running

We provide an example of estimating the ATE under the label-level privacy.

python main_label.py --dataset IHDP --N 5 --c 0.01 

Similarly, we provide an example of estimating the ATE under the sample-level privacy.

python main_sample.py --dataset IHDP --N 5 --h 0.001 --eps_1 0.1 --eps_2 0.7

Citation

@inproceedings{yuan2026private,
  title={{PrivATE: Differentially Private Average Treatment Effect Estimation for Observational Data}},
  author={Yuan, Quan and Li, Xiaochen and Du, Linkang and Chen, Min and Sun, Mingyang and Gao, Yunjun and He, Shibo and Chen, Jiming and Zhang, Zhikun},
  booktitle={NDSS},
  year={2026}
}