machine learning - GBM R function: get variable importance separately for each class -
i using gbm function in r (gbm package) fit stochastic gradient boosting models multiclass classification. trying obtain importance of each predictor separately each class, in picture hastie book (page 382).
however, function summary.gbm returns overall importance of predictors (their importance averaged on classes).
does know how relative importance values?
i think short answer on page 379, hastie mentions uses mart, appears available splus.
i agree gbm package doesn't seem allow seeing separate relative influence. if that's you're interested in mutliclass problem, pretty similar building one-vs-all gbm each of classes , getting importance measures each of models.
so classes a, b, c, & d. model vs. rest , importance model. model b vs. rest , importance model. etc.
Comments
Post a Comment