lee-group-cmu / RFCDE

Random Forests for Conditional Density Estimation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

default setting does not ignore irrelevant covariates

FridljDa opened this issue · comments

I manually added a line printing the covariate variable split during the training process (commit). In the official vigniette example from the package, there are 10 relevant and 10 irrelevant covariates. Running this vigniette and counting, how often relevant and irrelevant covariates are being split (commit), reveals, that they are being split equally often. The same problem also arose with other simulations with relevant and irrelevant covariates. Do you have ideas, how one might improve on this? Could a different parameter choice change this?

Screenshot 2022-03-04 at 13 50 31