Closed FlorianPargent closed 5 years ago
Hi, sorry for this, unfortunately mlr
broke the old feature selector names without doing something sensible like deprecating them first. I made some changes today that are related to this, can you check whether
library(mlrCPO)
pid.task %>>% cpoFilterFeatures(method = "FSelector_information.gain", perc = .5)
works now?
Thanks a lot for fixing it, it works now!
with the current developer version of mlr, neither the old nor the new filter methods work with mlrCPO:
Fehler in pid.task %>>% cpoFilterFeatures(method = "information.gain") : Assertion on 'method' failed: Must be element of set {'anova.test','auc','carscore','cforest.importance','FSelector_chi.squared','FSelector_gain.ratio','FSelector_information.gain','FSelector_oneR','FSelector_relief','FSelector_symmetrical.uncertainty','FSelectorRcpp_gain.ratio','FSelectorRcpp_information.gain','FSelectorRcpp_symmetrical.uncertainty','kruskal.test','linear.correlation','mrmr','permutation.importance','praznik_CMIM','praznik_DISR','praznik_JMI','praznik_JMIM','praznik_MIM','praznik_MRMR','praznik_NJMIM','randomForest.importance','randomForestSRC.rfsrc','randomForestSRC.var.select','ranger.impurity','ranger.permutation','rank.correlation','rf.importance','rf.min.depth','univariate','univariate.model.score','variance'}, but is 'information.gain'.
Fehler in checkParamsFeasible(par.set, present.pars) : FSelector_in... is not feasible for parameter 'method'!