Priv-PC: Differentially Private Causal Graph Discovery

This is the code accompanying the Neurips 2020 paper "Towards practical differentially private causal graph discovery". The code for the original pc algorithm is borrowed from this excellent repo.

Overview

Priv-PC is designed for differentially private causal graph discovery. Priv-PC leverages sieve-and-examine mechanism to augment PC algorithm with differential privacy. Intuitively, Priv-PC uses sparse vector technique to sieve out "unsignificant" queries while using substantial privacy budget to carefully examine "significant" ones.

Prequisities

• Python 3.6.10
• R 3.4.4

Reproduce the evaluation results

First, download all dependencies by running pip install -r requirements.txt.

The evaluation can be reproduced using python eval.py name_of_dataset.