search

imbs-hl / ranger

A Fast Implementation of Random Forests
http://imbs-hl.github.io/ranger/
774 stars 194 forks source link

Feature Request: Bagging with a randomized split rule #347

Closed wbnicholson closed 6 years ago

wbnicholson commented 6 years ago

While I'm aware that the original extratrees algorithm uses the entire data set in every tree, I think that it would be a logical extension for ranger to allow the combination of bootstrapping and discretization of the parameter split space (i.e. the option num.random.splits combined with "variance" splitrule). This would be similar to the functionality in randomForestSRC with the nsplit parameter.

Would this be a straightforward extension?

mnwright commented 6 years ago

Actually that is the way it's implemented in ranger. We are using bagging and random subspace in nodes and the underlying splitrule in "extratrees" for regression is "variance".

wbnicholson commented 6 years ago

Great, thanks for the clarification.