mlr3 learner train / predict timeout

[mb706]

When using "callr" encapsulation for learners, callr::r is used so we should in be able to have a training / prediction timeout for that learner. This would be necessary to make mlr3automl (which @a-hanf is working on) more robust against timeouts / stalling Learners.

[RaphaelS1]

I'm not sure if this is related or a separate issue, but a time-out feature could also help in situations such as #501 where benchmarking is crashing due to conflicting parallelisation methods

[mllg]

Timeout now supported in mlr3misc's invoke() and encapsulate(). I will implement it here once mlr3misc is on CRAN.

[RaphaelS1]

Timeout now supported in mlr3misc's invoke() and encapsulate(). I will implement it here once mlr3misc is on CRAN.

Will this detect timeouts due to parallelisation? I know the logger doesn't handle this well which is why I'm checking

[jakob-r]

@mllg What's the current status?

[mllg]

It is possible to set learner$timeout. Not yet possible on the resample/benchmark level (#576).