mCMIkNN

A kNN-based Non-Parametric Conditional Independence Test for Mixed Data and Application in Causal Discovery

This repository contains the code, the experiments, and the Appendix of the paper "A kNN-based Non-Parametric Conditional Independence Test for Mixed Data and Application in Causal Discovery" which is accepted at the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases ECML PKDD 23.

Setup and Example

To setup the code execute the following command.

git clone [email protected]:hpi-epic/mCMIkNN.git
cd src
pip3 install -r requirements.txt

You can also install the independence tests via

pip3 install .

which lets you later call

import indeptests

Note, within the requirements, we added manm-cs only to allow rerunning some of the code for the experiments and results shown in the associated publication. This is not required for execution of the conditional independence test mCMIkNN.

After setup, you can execute the following command as an example. python3 main.py -i ../data/coolingData_1000.csv

The main.py script takes the following parameter as input:

Parameter	Default	Description
-i	-	Input File location, with column description in header
-l	-	Maximum size of separation set
-c	2	Number of cores to execute tests in parallel
-p	100	Number of permutations $M_{perm}$
-t	-	Optional transformation of data. Options are "normalize", "rank", "standardize", "uniform" default is none.
-k	25	Value $k_{CMI}$ of knn in cmi estimation
-r	5	Value $K_{perm}$ of knn in local permutation scheme
-a	0.01	Nominal value $\alpha$, i.e., significance level

Contributor

• Daniel Thevessen
• Johannes Huegle
• Christopher Hagedorn