About P3GM

P3GM is a differentially private generative model. The algorithm of P3GM is described in our paper (https://arxiv.org/abs/2006.12101).

How to Run

This code is implemented by python3.7.9.

Install P3GM.

git clone https://github.com/tkgsn/P3GM
cd P3GM

Install tensorflow_privacy library.

git clone https://github.com/tensorflow/privacy.git

Make a virtual enviroment and install the needed dependencies.

python -m venv venv
. venv/bin/activate
pip install -r requirements.txt

Download the datasets.

./download_dataset.sh

Run the experiments.

./exp.sh

Datasets

Kaggle Credit Dal Pozzolo, Andrea, et al. "Calibrating probability with undersampling for unbalanced classification." 2015 IEEE Symposium Series on Computational Intelligence. IEEE, 2015.

Isolet: https://archive.ics.uci.edu/ml/datasets/isolet

ESR: https://archive.ics.uci.edu/ml/datasets/Epileptic+Seizure+Recognition

Adult: https://archive.ics.uci.edu/ml/datasets/adult