[SOLVED] Predict on holdout set using mlr3

Predict on holdout set using mlr3

I am using mlr3 package. I set roles of some rows to "holdout" and than trained the model:

library(mlr3)

# train on iris
task = tsk("iris")
task$nrow
task$set_row_roles(130:150, "holdout")
learner = lrn("classif.rpart")
learner$train(task)

How can I know use holdout set to make prediction on them?

# predict on holdout
task$row_roles$holdout
## HOW TO PREDICT ON HOLDOUT SET ? 
# learner$predict()

Solution

You can set the row ids that you want to get the prediction for in the $predict() method.

library(mlr3)

task = tsk("iris")
task$nrow
#> [1] 150
task$set_row_roles(130:150, "holdout")
learner = lrn("classif.rpart")
learner$train(task)

learner$predict(task, row_ids = task$row_roles$holdout)
#> <PredictionClassif> for 21 observations:
#>     row_ids     truth   response
#>         130 virginica versicolor
#>         131 virginica  virginica
#>         132 virginica  virginica
#> ---                             
#>         148 virginica  virginica
#>         149 virginica  virginica
#>         150 virginica  virginica

^{Created on 2023-02-10 with reprex v2.0.2}