Erik Sverdrup

Holding on a bit on this since RStudio is tightly integrated with testthat and RStudio shortcuts is convenient to have.

Hi @ZhiyLiu, policytree is mainly intended for building shallow trees on moderately sized data. Depth 4 is something that would only be computationally feasible on small data, or data with...

@ZhiyLiu the new feature branch #118 has a method to build deeper (i.e depth >= 3) trees by performing "locally" optimal `search.depth` splits.

Rcpp::checkUserInterrupt() was not added since the solver was supposed to be language agnostic, but we can add it with #ifdef's for R at compile time if there is interest. If...

Hi @njawadekar > For example, would it be reasonable to develop a protocol whereby we only input covariates into the policy tree that were listed among the top 10% of...

Also, if you're looking for a real-world empirical application, @hhsievertsen has a paper using causal forest + policy tree here https://github.com/hhsievertsen/hhsievertsen.github.io/raw/master/mat/wp/chx_sep2021.pdf

There is a note on scaling in the online documentation here: https://grf-labs.github.io/policytree/articles/policytree.html#gauging-the-runtime-of-tree-search As you see the cardinality of the the Xj's is important, and you can speed things up by...

For practical reference, here is a short table of empirical run times for `policy_tree` (version 1.0). | depth | n (continuous) | features | actions | split.step | time |...

That depends on what exactly what you are trying to do. You could for example estimate CATE on a training set and evaluate [RATE](https://grf-labs.github.io/grf/articles/rate.html) on a test set and to...

PS: @yqi3 if you are trying out GRF in a [rdd setting](https://github.com/yqi3/Replication-GRF-RD/blob/84f3782a1242e397c609771867c2604742e91a96/Replications/Tables/Table%20A.1%20(Toy%20Sim)/(2)%20RF/Sim_RF_1d.R#L39) the new estimator `lm_forest` (grf version 2.2.0+) can be used for this purpose, as a local linear regression...