selected_features for learners that don't support it should be the entirety of features seen in training

[mb706]

This way we could correctly query a pipeline that selects features first and gives the result to a learner. The GraphLearner could then ask the learner at the end how many features it used, and if it is a learner that supports embedded featsel (rpart e.g.) then this would give the correct value, but even for learners that do not do support it the result could make sense.

Also this would solve mlr-org/mlr3fselect#87

[be-marc]

library(mlr3)
library(mlr3learners)

learner = lrn("classif.rpart")
task = tsk("spam")

learner$train(task)
learner$selected_features()

#> [1] "charDollar"      "hp"             
#> [3] "remove"          "charExclamation"
#> [5] "capitalTotal"    "free"  

learner = lrn("classif.log_reg")
learner$train(task)
learner$selected_features()
# > Error: attempt to apply non-function

[berndbischl]

so first order of business would be here to extend the docs, the docs don't say what happens if the property does not exists

[mb706]

currently mlr3pipelines handles this on its own end if this flag is set:

https://github.com/mlr-org/mlr3pipelines/blob/b1042d7967d13207276f6c1e429dfca86c76416f/R/GraphLearner.R#L73-L84

[berndbischl]

i would suggest this
a) the learner has an member var / option:
selected_features_not_supported = "error" / "all"
this controls what happens in "selected_features"

b) selected_features as a method is present in all learners.
by default it return an error (if not implemented)

[mb706]

selected_features_impute

[berndbischl]

we also need to remove this from pipelines then

[be-marc]

Closed by #1230