daniel sousa - 2015-01-21

First of all, congratulation for RankLib, which is a power tool.

I'm working with RandomForest, and as I have read in RankLib Forums, the
RandomForest were implemented for regression.

However, the default value to the parameter -rtype is MART( a.k.a. Gradient
boosted regression tree), then, I would like to ask if is each tree in
RandomForest build either as CART algorithm, or using the Gradient Bossted
Regression Tree (GBRT)?

I guess that using GBRT will conduct to a better accuracy than CART
algorithm. But, I guess this is a modification from original
RandomForest Algorithm.

Thanks for all

 

Last edit: daniel sousa 2015-01-22