Variable importance doesn't work with blackboost models + possibly wrong output of blackboost$xselect()

[hbaniecki]

I found out that varimp() doesn't work with blackboost models.

Code to reproduce the error:

library(mboost)
library(survival)
model <- blackboost(time ~., data=veteran)
vi <- varimp(model)
plot(vi)

I wanted to fix it, but it got complicated, see

mboost/R/varimp.R

Lines 8 to 10 in 739cbe9

	learner_names <- if( !(inherits(object, "glmboost")) )
	names(object$baselearner) else variable.names(object)
	learner_selected <- object$xselect()

If learner_names are meant to be variable names, then we could do

if (inherits(object, "blackboost")) {
    learner_names <- object$baselearner[[1]]$get_names()
}

But then, object$xselect() returns a vector of only ones, should this be the case?

model$xselect()

  [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
 [39] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
 [77] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

With glmboost, everything seems to work fine?

model <- glmboost(time ~ ., data = veteran)
vi <- varimp(model)
model$xselect()
 [1]  7  7  7  7  7  7  7  3  4  3  7  4  3  4  3  4  7  3  4  3  4  7  3  4  3
[26]  4  7  3  4  3  4  7  3  4  3  4  7  3  4  7 10  4  3  6  4  3  7  6  4  3
[51] 10  4  6  7  3  4 10  6  7  3  4  6  3  4  2 10  7  6  2  4  3  7 10  6  5
[76]  4  3  5  4  3  2  7  6  5  4  3  5  4 10  6  7  3  5  4  2  3  5  4  6  7
plot(vi)