FlexCD

FlexCD is a novel method for causal discovery using Gaussian Processes in the context of additive noise models. The method introduces a simple Gaussian Process that is comprised of a mean function accounting for linearity, a homoscedastic Gaussian Process component accounting for static levels of noise and a heteroscedastic Gaussian Process component accounting for changing levels of noise. These are optimised to model the causal generative process. The interplay of each of these three components allows FlexCD to identify cause and effect without making too strong assumptions with regards to the nature of the true underlying causal function and noise. The package needs to have FastICA installed.
Download