Erik Sverdrup

Ok, you could use grf's regression_forest to estimate P[Y = 1 | V, Z] then plug in and divide. I.e. `rf = regression_forest(cbind(V, Z), Y); p.hat.z0 = predict(rf, cbind(X, z0))`

> Thanks @swager > Is there a reason that the best pruned tree in the forest is not merged into the master branch yet? @ginward the best pruned tree requires...

While GATES/CLAN looks like an interesting addition, we have no immediate plans of turning it into a feature. A "GRF native" alternative which can serve a similar purpose is RATE:...

Hi @beniaminogreen, thanks a lot, this looks very cool and interesting! I don't have time to look closely until a bit later next month, but in the meantime, could I...

Thanks Ben, that looks great. This is very impressive work. I’m trying to understand where the gains are coming from, could you please run that benchmark plot for dense Xj,...

Sorry for getting you to run this, but I only have my phone at the moment. Could you please run that same script but overlay a 3rd line with runtime...

Thanks, yeah that is why I wanted to see the depth 0 timing, had a hunch a big speed difference arose from the sorted sets being created faster in Rust...

Great, thanks! Could you please run similar benchmarks for a depth 2 tree on three X-configurations: {all binary, a handful of discrete values, dense (i.e `runif(n*p)`)}? (policytree may be to...

Thanks! I'd like a larger number of observations to see what's going on, could you do up to n = 100k, at least for the top plot? It could take...

Actually, you could leave out policytree from the benchmarks, we have roughly an idea of how long time that takes if you include an R file with benchmark details. What...