we need to document what happens in cases of learner error

[mlr-org]

we need to document what happens in cases of learner error

[berndbischl]

our mechanism should probably be:
run with callr in an external process (user has to select this), then we capture the fail.
if that happens, we can use a dummy predictor.
if the user does that, he needs to be able to transparently see (at the end) that that has happened and due to what error

[mllg]

There is a vignette now. We have a fallback learner which you can define per learner. Calls to learners are encapsulated via evaluate or callr (more overhead, but guards against segfaults) if enabled via mlr_control(). Corresponding logs are stored in the experiment.

Default: Exception is raised, no error handling.