I am currently working on a very large dataset (billions of rows) of A/B test data and want to implement some methods to estimate conditional average treatment effects. I basically need a forecast what the impact of the binary A/B tested switch is on an individual depending on his covariates/features.
I found the econml package which is a tremendous resource in that regard, implementing honest forests, X/S/T-learner etc., but as far as I see none of these methods allow for an incremental training like e.g. partial_fit in sklearn. Are you aware of any implementations of these or similar algorithms that can be trained incrementally?