Erik Sverdrup

Hi @Jinwoo-Yi, it sounds like with just 49 W1=1 then lack of overlap could potentially be an issue: https://grf-labs.github.io/grf/articles/diagnostics.html#assessing-overlap. Instead of fitting two forests, you could also use `multi_arm_causal_forest` with...

1) https://bookdown.org/halflearned/tutorial/pe1.html#evaluating-policies-in-observational-settings 2) You could estimate policy on different folds as in section 5.1 in https://arxiv.org/pdf/1702.02896.pdf

There's no universal guideline, a 50/50 train/test split is just a reasonable default. And for different splits, that's normal considering trees are discontinuous. Ideally, if there is signal, the estimated...

Hi @njawadekar, Sorry, policy_tree currently does not support missing X's. You could fit it using the non-missing units. ``` subset

Hi @Matthew-R-Lehnert, this vignette https://grf-labs.github.io/grf/articles/muhats.html has an example of how to recover the implicit E[Y(1)|X] and E[Y(0)|X] estimates from, E[Y|X], E[W|X] and the CATEs

For reference [here](https://github.com/erikcs/pygrf) is an old Cython wrapping of some basic functionality in grf v 0.10.2, but as @halflearned mentions it is easier to just use rpy2.

causal_forest estimates a difference in conditional means E[Y(1) - Y(0) | X], a mean is sensitive to outliers, so yes, causal_forest estimates can be sensitive to large outliers in Y....

High @minhengw, you could try and plot a histogram of the instrument propensities (`forest$Z.hat`) and see if there's a region where they are far from 0 or 1 then pass...

> > High @minhengw, you could try and plot a histogram of the instrument propensities (`forest$Z.hat`) and see if there's a region where they are far from 0 or 1...

Hi @tibshirani, I think @jtibshirani and @swager may have intentionally left that out to keep affinity with the [GRF theory](https://grf-labs.github.io/grf/REFERENCE.html#mtry): >[](https://grf-labs.github.io/grf/REFERENCE.html#minnodesize)To more closely match the theory in the GRF paper,...