causalnex
causalnex copied to clipboard
Relationship between do intervention versus uplifting models based with causal graph aware estimand
I've been browsing through the codebases of several popular causal inference libraries and I am especially interested in the relationship of do-intervension
vs uplifting models (methods in econml/causalml).
So here is my understanding:
It feels like do-intervention
does a pretty similar thing as compared to doWhy
's identify-estimate step, where the estimation of an action is evaluated on an estimand retrieved from a causal graph with do calculus, but the only difference is causalnex is using bayesian network for inference instead of models commonly used in CATE.
Am I thinking it in a correct way or did I miss anything? If they are different, what are their particular use cases and what are the caveats I should be aware of when choosing these methods?