Re: Equivalent hyperparameter value for “auto“, ”sqrt”, “log2” and “None” (SCIKIT) in weka?
RandomForest only has the log2 heuristic (which is the default). However, you
can specify the exact number of attributes to use instead.
In practice, it is probably be best to tune the value for a particular
dataset anyway, using internal cross-validation. You can do that in WEKA
using CVParameterSelection (the old approach) or MultiSearch (the more
flexible, newer approach). Wrap RandomForest into one of those two meta