Learning with confounders variables about causalnex HOT 4 CLOSED

Hello and thanks for your question.
The algorithm is valid under the no-unobserved-confounder assumption.

Regarding alternatives, do you have something specific in mind?

Philip

from causalnex.

Hey @arita37 thanks for your question. I wanted to follow up from @qbphilip reply and see if you are aware of any other packages / publications / algorithms that offer interesting alternatives? This is definitely something we could explore.

Has this no-unobserved-confounder assumption caused you issues when using CausalNex? If so, are you able to give us any insights into them? Perhaps this could also help us look into potential solutions.

Ben

from causalnex.

I guess if you could add unobserved variables (or is this actually possible?), backdoor adjustment might be an option. See e.g. pgmpy or DoWhy.
@benhorsburgh @qbphilip would you be interested in importing such features? I think it may be of great value. I'd be interested in contributing.

from causalnex.

A new feature to add a latent variable and learn its CPD using EM algorithm has been implemented in this commit and will be made available in the next release.

Additionally, a new tutorial has been created to showcase how to utilise this feature to learn with confounding variables.

Closing this issue for now, but feel free to open a new one if you have any problem/questions regarding the new feature.

from causalnex.