Feature selection - "argument 'method' matched by multiple actual arguments" error when tuning with mlrMBO

[notiv]

I'm using mlrMBO to tune an xgboost model with a few wrappers. After adding the following wrapper:

lrn <- mlr::makeFilterWrapper(lrn, method = 'ranger.permutation')

I get the following error:

Error in (function (task, method = "randomForestSRC.rfsrc", fval = NULL,  : 
  formal argument "method" matched by multiple actual arguments 

This issue looks similar to the following resolved issue: https://github.com/mlr-org/mlr/issues/1066

Any idea what could be wrong?

Session Info (part): R 3.5.1 mlrMBO_1.1.1
mlr_2.12.1
ParamHelpers_1.10

P.S. If a reproducible example could be useful, I'll try to provide one. I won't be sad, if we can get away without one ;-)

[larskotthoff]

A reproducible example would be great...

[notiv]

Ok, here you go (not all steps and values make sense for the titanic dataset, but you get the point):

library(mlrMBO)
library(dplyr)
library(titanic)

data <- titanic_train
sample <- sample.int(n = nrow(data), size = floor(.7*nrow(data)), replace = F)

train <- titanic_train[sample, ] %>% select(Pclass, Sex, Age, SibSp, Fare, Survived) %>% mutate(Sex = ifelse(Sex == 'male', 0, 1))

mlr::configureMlr(on.par.without.desc = "quiet")
lrn <- mlr::makeLearner(cl = 'classif.xgboost',
                        predict.type = "prob",
                        fix.factors.prediction = TRUE,
                        tree_method = 'exact')

lrn <- mlr::makeFilterWrapper(lrn, method = 'chi-squared')

lrn <- mlr::makeImputeWrapper(lrn,
                              classes = list(integer = mlr::imputeMedian(),
                                             numeric = mlr::imputeHist(),
                                             factor = mlr::imputeMode()),
                              dummy.classes = "factor")

classif.task <- mlr::makeClassifTask(data = train,
                                     target = "Survived",
                                     positive = "1")

eval_metrics <- list(auc, fpr, tpr, ppv, f1, acc, ber, mmce, timetrain)

param_set <- ParamHelpers::makeParamSet(
  ParamHelpers::makeIntegerParam(id = "nrounds", lower = 10, upper = 100, default = 20),
  ParamHelpers::makeNumericParam(id = "eta", lower = -7, upper = -5, default = -6, trafo = function(x) 2^x),
  ParamHelpers::makeIntegerParam(id = "max_depth", lower = 3, upper = 5, default = 4),
  ParamHelpers::makeNumericParam(id = "colsample_bytree", lower = 0.4, upper = 0.9, default = 0.6),
  ParamHelpers::makeNumericParam(id = "subsample", lower = 0.4, upper = 0.9, default = 0.5),
  ParamHelpers::makeIntegerParam(id = "fw.abs", lower = 5, upper = 8, default = 6)
)

ctrl = makeMBOControl()
ctrl = setMBOControlTermination(ctrl, iters = 5)
tune.ctrl = makeTuneControlMBO(mbo.control = ctrl)

resInst <- mlr::makeResampleInstance("CV", iters = 4, task = classif.task)

res_ext <- mlr::tuneParams(lrn,
                           classif.task,
                           resInst,
                           par.set = param_set,
                           control = tune.ctrl,
                           measures = eval_metrics,
                           show.info = TRUE)

And here my (slightly different to the previously mentioned) sessionInfo:

other attached packages:
[1] mlrMBO_1.1.1      smoof_1.5.1       checkmate_1.8.5   BBmisc_1.11       bindrcpp_0.2.2    titanic_0.1.0     dplyr_0.7.5      
[8] mlr_2.12.1        ParamHelpers_1.10

loaded via a namespace (and not attached):
 [1] RWeka_0.4-38          tidyselect_0.2.4      DiceKriging_1.5.5     purrr_0.2.5           splines_3.5.0        
 [6] rJava_0.9-10          lattice_0.20-35       colorspace_1.3-2      htmltools_0.3.6       viridisLite_0.3.0    
[11] yaml_2.1.19           FSelector_0.31        survival_2.41-3       XML_3.98-1.11         plotly_4.7.1         
[16] rlang_0.2.1           pillar_1.2.3          glue_1.2.0            entropy_1.2.1         xgboost_0.71.2       
[21] plot3D_1.1.1          RColorBrewer_1.1-2    lhs_0.16              mco_1.0-15.1          bindr_0.1.1          
[26] plyr_1.8.4            munsell_0.4.3         gtable_0.2.0          htmlwidgets_1.2       misc3d_0.8-4         
[31] RWekajars_3.9.2-1     parallelMap_1.3       parallel_3.5.0        Rcpp_0.12.17          scales_0.5.0         
[36] backports_1.1.2       randomForestSRC_2.6.1 jsonlite_1.5          ggplot2_2.2.1         packrat_0.4.9-3      
[41] digest_0.6.15         stringi_1.2.2         RJSONIO_1.3-0         grid_3.5.0            tools_3.5.0          
[46] magrittr_1.5          lazyeval_0.2.1        tibble_1.4.2          randomForest_4.6-14   tidyr_0.8.1          
[51] pkgconfig_2.0.1       Matrix_1.2-14         data.table_1.11.4     assertthat_0.2.0      httr_1.3.1           
[56] R6_2.2.2              compiler_3.5.0       

[larskotthoff]

Thanks. The problem is down to the method argument you're giving when constructing the filter wrapper:

lrn <- mlr::makeFilterWrapper(lrn, method = 'chi-squared')

As far as I can tell, this argument does nothing (the filter wrapper doesn't have an argument like that, and the random forest used underneath doesn't have either). Removing it fixes the error for me.

[notiv]

Ah, stupid me... I got the parameters from the filterFeatures function. Just in case someone makes the same mistake, the correct call would be:

lrn <- mlr::makeFilterWrapper(lrn, fw.method = 'chi.squared')

Thanks Lars and sorry for wasting your time :-/