maybe we could add a parameter to flag the variables that are discrete like the mutual_info_classif from sklearn