update.train making .weights an independent variable when method = 'ranger'

[mb158127]

As the title says. I am trying to override the train object $bestTune to get a different $finalModel for prediction. I am not entirely sure if this should be placed here or with ranger. Thanks.

Minimal, reproducible example:

library(caret)

grid <- expand.grid(mtry = 1:2,
                    min.node.size = c(5, 10, 15),
                    splitrule = "variance" )

ctrl <- trainControl(method = "cv",
                     number = 5)

## define response for model
response <- iris$Sepal.Length / iris$Sepal.Width
## define case.weights for ranger
wgt <- iris$Sepal.Width

## fit initial train using ranger
trn.ranger <- train( y = response,
                     x = iris[, 3:5],
                     weights = wgt,
                     method = "ranger",
                     tuneGrid = grid,
                     verbose = F,
                     trControl = ctrl,
                     num.trees = 20,
                     importance = "impurity")

varImp(trn.ranger) ## 3 features in model
trn.ranger$bestTune ##bestTune is mtry=1, min.node.size = 10

## update to choose different bestTune
trn.ranger.update <- update(trn.ranger, param = list(mtry = 2, 
                                                     splitrule = "variance", 
                                                     min.node.size = 15))
varImp(trn.ranger.update) ## new 4th feature is .weights

## error due to .weights column does not exist in newdata
predict(trn.ranger.update, newdata = iris)

Session Info:

> sessionInfo()
R version 3.5.0 (2018-04-23)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows Server >= 2012 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252   
[3] LC_MONETARY=English_United States.1252 LC_NUMERIC=C                          
[5] LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] caret_6.0-80         ggplot2_3.0.0        lattice_0.20-35      RevoUtils_11.0.0    
[5] RevoUtilsMath_11.0.0

loaded via a namespace (and not attached):
 [1] Rcpp_0.12.17        lubridate_1.7.4     tidyr_0.8.1         class_7.3-14        assertthat_0.2.0   
 [6] ipred_0.9-6         psych_1.8.4         foreach_1.5.0       ranger_0.9.0        R6_2.2.2           
[11] plyr_1.8.4          magic_1.5-8         stats4_3.5.0        e1071_1.6-8         pillar_1.2.3       
[16] rlang_0.2.1         lazyeval_0.2.1      kernlab_0.9-26      rpart_4.1-13        Matrix_1.2-14      
[21] splines_3.5.0       CVST_0.2-2          ddalpha_1.3.3       gower_0.1.2         stringr_1.3.1      
[26] foreign_0.8-70      munsell_0.5.0       broom_0.4.4         compiler_3.5.0      pkgconfig_2.0.1    
[31] mnormt_1.5-5        dimRed_0.1.0        nnet_7.3-12         tidyselect_0.2.4    tibble_1.4.2       
[36] prodlim_2018.04.18  DRR_0.0.3           codetools_0.2-15    RcppRoll_0.2.2      dplyr_0.7.6        
[41] withr_2.1.2         MASS_7.3-49         recipes_0.1.2       ModelMetrics_1.1.0  grid_3.5.0         
[46] nlme_3.1-137        gtable_0.2.0        magrittr_1.5        scales_0.5.0        stringi_1.2.2      
[51] reshape2_1.4.3      bindrcpp_0.2.2.9000 timeDate_3043.102   robustbase_0.93-0   geometry_0.3-6     
[56] lava_1.6.1          iterators_1.0.10    tools_3.5.0         glue_1.2.0          DEoptimR_1.0-8     
[61] purrr_0.2.5         sfsmisc_1.1-2       abind_1.4-5         parallel_3.5.0      survival_2.41-3    
[66] yaml_2.1.19         colorspace_1.3-2    bindr_0.1.1  

[mb158127]

Please let me know if there is something more I should include @topepo

[topepo]

Sorry, just now rotating back to caret.

It is fixed but, to be honest, I'll be merging PRs and fixing bugs until the weekend. I would wait until then before testing.

library(caret)
#> Loading required package: lattice
#> Loading required package: ggplot2

grid <- expand.grid(mtry = 1:2,
                    min.node.size = c(5, 10, 15),
                    splitrule = "variance" )

ctrl <- trainControl(method = "cv",
                     number = 5)

## define response for model
response <- iris$Sepal.Length / iris$Sepal.Width
## define case.weights for ranger
wgt <- iris$Sepal.Width

## fit initial train using ranger
trn.ranger <- train( y = response,
                     x = iris[, 3:5],
                     weights = wgt,
                     method = "ranger",
                     tuneGrid = grid,
                     verbose = F,
                     trControl = ctrl,
                     num.trees = 20,
                     importance = "impurity")

varImp(trn.ranger) ## 3 features in model
#> ranger variable importance
#> 
#>              Overall
#> Petal.Width   100.00
#> Species        40.53
#> Petal.Length    0.00
trn.ranger$bestTune ##bestTune is mtry=1, min.node.size = 10
#>   mtry splitrule min.node.size
#> 3    1  variance            15

## update to choose different bestTune
trn.ranger.update <- update(trn.ranger, param = list(mtry = 2, 
                                                     splitrule = "variance", 
                                                     min.node.size = 15))
varImp(trn.ranger.update) ## new 4th feature is _no longer_ .weights
#> ranger variable importance
#> 
#>              Overall
#> Petal.Length  100.00
#> Petal.Width    61.78
#> Species         0.00

Created on 2018-11-14 by the reprex package (v0.2.1)