gcit

Conditional Independence Testing with Generative Adversarial Networks

This is a python implementation of the algorithm in the paper "Conditional Independence Testing with Generative Adversarial Networks". The goal of this project is to test conditional independence between variable sets X and Y conditional on Z. It contains synthetic data generation to understand the behaviour of our test in various settings.

Please cite the above paper if this resource is used in any publication

Dependencies

The only significant dependencies are python 3.6 or later. It is known to work with tensorflow 1.13.1.

First steps

To get started, check tutorial_gcit.ipynb which will guide you through the test from the beginning.

Use case on genetic data

We include in the CCLE Experiments folder the code used in the real data experiment on Section 5 of the main body of this paper. The folder includes the data used and a simple script to test conditional independence of each feature and drug response given all other features.