fallback learner should maybe be a warning

[sebffischer]

Currently, the fallback learner log info is only shown during debug mode:

mlr3/R/worker.R

Lines 94 to 102 in dc2a983

	lg$debug("Calling train method of fallback '%s' on task '%s' with %i observations",
	fb$id, task$id, task$nrow, learner = fb$clone())
	fb = assert_learner(as_learner(fb))
	require_namespaces(fb$packages)
	fb$train(task)
	learner$state$fallback_state = fb$state
	lg$debug("Fitted fallback learner '%s'",

maybe this should be a warning?

[mb706]

Maybe relates more to the 'encapsulate' machinery than the fallback learner -- what to do if something that was encapsulated fails? Or is there a way for the fallback learner to be invoked without the encapsulation noticing?

[sebffischer]

make this info

[be-marc]

learner = lrn("classif.debug", error_train = 1, fallback = lrn("classif.featureless"))

learner$train(tsk("spam"))