Not sure best approach yet, needs to be explored. We could do a grid search over mtry and nfeatures, or we could just train mtry, and then after picking it, tune nfeatures, but these two parameters interact heavily so probably better to do grid search with (2, 5, 10, 25, 50, 100... features) (and it will be very slow).
Not sure best approach yet, needs to be explored. We could do a grid search over mtry and nfeatures, or we could just train mtry, and then after picking it, tune nfeatures, but these two parameters interact heavily so probably better to do grid search with (2, 5, 10, 25, 50, 100... features) (and it will be very slow).