Repository navigation
Breaking changes
-
bag_tree()with the"rpart"engine no longer accepts case weights. Fit the model without case weights instead (#390). -
bag_tree()with the"rpart"engine now passestree_depth,min_n, andcost_complexityon to the trees, so default fits now use thebag_tree()defaults ofcost_complexity = 0andmin_n = 2. To restore the previous fits, setcost_complexity = 0.01andmin_n = 20(#406). -
boost_tree()with the"mboost"engine no longer accepts case weights because mboost doesn't predict from a weighted fit. Fit the model without case weights instead (#363). -
boost_tree()with the"mboost"engine no longer allows settingnuviaset_engine(). Uselearn_rateinstead (#395). -
boost_tree()with the"mboost"engine now warns and ignorescontrolandtree_controlswhen set viaset_engine(). Previously they silently overrode the main arguments, such astreesandtree_depth. Use the main arguments ofboost_tree()instead (#414). -
survival_prob_*()functions now error when given the defuncttimeargument. Useeval_timeinstead (#419). -
survival_prob_coxph()no longer has the defunctxargument. Useobjectinstead (#419).
New features
-
boost_tree()with the"mboost"engine now passeslearn_rateon tomboost::boost_control(nu), making it tunable. Previously it was silently ignored (#395). -
multi_predict()is now available forboost_tree()with the"mboost"engine over thetreessubmodel parameter (#290). -
null_model()gains a"survival"engine for censored regression, fitting a Kaplan-Meier curve viasurvival::survfit()(#353). -
rand_forest()gains a"randomForestSRC"engine for censored regression, fitting a survival random forest viarandomForestSRC::rfsrc()(#130). -
rand_forest()gains a"ranger"engine for censored regression, fitting a survival random forest viaranger::ranger()(#131).
Minor improvements and fixes
-
bag_tree()with the"rpart"engine now passestree_depth,min_n, andcost_complexity, as well as otherrpart::rpart.control()arguments set viaset_engine(), on to the trees. Previously they were silently ignored (#406). -
decision_tree()with the"rpart"engine now correctly returns the median survival time of the leaf's Kaplan-Meier curve fortype = "time"predictions, instead of rpart's relative event rate (#331). -
proportional_hazards()with the"glmnet"engine now predicts on data with factors when fitted throughfit_xy()(#365). -
proportional_hazards()with the"survival"engine now predicts when a singlestrata()term contains more than one variable, e.g.strata(s1, s2)(#404). -
survival_prob_*()andhazard_*()now validate their inputs and return more informative error messages when given an unusableobject,new_data, oreval_time(#271). -
survival_reg()with the"survival"engine andstrata()terms now returnsNAsurvival and hazard predictions for new data rows with a missing value in a stratification variable, instead of silently using another stratum's scale (#383). -
Engine-specific tuning parameters are now registered by censored instead of parsnip (#417).