James Jong
2013-Feb-12 20:10 UTC
[R] Putting priors on which factors to sample more from, in random forests
I was wondering if anyone knows of a random forest implementation (or way of tweaking the standard randomForest library) that allows one to specify some sort of variable importance a* *priori. For example, say that I know that some variables/factors could be more informative than others for classification. I would like to help RF look at these variables more than others during splitting. As far as I know, the standard implementations choose factors fully randomly (i.e. uniform distribution) at each node. Is there a way to influence this selection? Thanks! James [[alternative HTML version deleted]]