Pairwise Causal Discovery with Support Measure Machines

This repo contains the code to replicate the experiments in the manuscript Pairwise Causal Discovery with Support Measure Machines

requirements

Experiments are run on python v3.8.12. We use the framework for pairwise causal discovery implemented in the The Causal Discovery Toolbox. Our method relies on kernel computed using JAX and SVM available through the sklearn wrappers of libsvm.

The complete list of required python packages is in requirements.txt.

Additionally, the scripts to generate figures are written in R and requires the ggplot2, ggridge and reshape2 packages.

SMM weighted ensemble

We implement our proposed method in the SMMwEnsemble class available in smmw_ensemble/ensemble.py. An object of this class is instantiated with the following code:

from smmw_ensemble import SMMwEnsemble
from cdt.causality.pairwise import CDS, ANM

model = SMMwEnsemble({"CDS": CDS(), "ANM": ANM()},
                     gamma = 10)