subsample columns

[mayer79]

Hello Beth

Awesome package.

There are very efficient random forest packages out there, but still I'd love to see the option "colsample = 1" in rpart.control. If set to a value below 1, then in each split, a random sample of size

1 <= ceiling(m * colsample) <= m

would be drawn among the $m$ predictors, like in a random forest.

There are two reasons for this:

1. For teaching, it would be a fantastic way to program a simple random forest based on rpart.

2. It would allow to make a random forest with Poisson split rule, something I am waiting for since long.