When training a typical random forest model, there are two primary sources of randomness: bagging and selecting features to consider at each split. Previously we did not pass the rng to the splitter, so the latter was not reproducible. This PR fixes that.
When training a typical random forest model, there are two primary sources of randomness: bagging and selecting features to consider at each split. Previously we did not pass the
rngto the splitter, so the latter was not reproducible. This PR fixes that.