forum.alglib.net

on the ALGLIB description page for random forest, it says:

"The algorithm contains three parameters that need to be adjusted: the ratio r - size of the part of the training set which will be used for the construction of individual trees; the number of trees NTrees; as well as the number NFeatures indicating the number of variables used to grow individual trees."

But the dfbuildrandomdecisionforest function does not let us specify the NFeatures parameter. Is there another place
we can set it? or the implementation just automatically sets NFeatures to (total # of features)/2?
or ....? Thanks.

Yes, you are right. Implementation automatically uses NVars/2 as number of features. Development version of ALGLIB random forests allowed to tune this parameter, but in the final release this feature was hidden. I'll change documentation on the next site update.