add reduceResultsList.fun arg to reduceResultsBatchmark

[tdhock]

Hi! in my use case, I want to do model interpretation, so I am using store_models=TRUE.
But that makes my cluster system kill my R process (too much RAM used) when I do reduceResultsBatchmark, which gets killed when it calls reduceResultsList, #18 (comment) because the total of all learner_state is too large.
So in this PR I add a new argument, reduceResultsList.fun, which gets passed on, so then I can process the learner_state. For example the code below works for me using this PR. (uses reasonable amount of memory / R is not killed)

ignore.learner <- function(L){
  L$learner_state$model <- NULL
  L
}
bmr = mlr3batchmark::reduceResultsBatchmark(ids, reg = reg, store_backends = FALSE, reduceResultsList.fun=ignore.learner)

(but without the proposed reduceResultsList.fun, the above code results in my cluster system killing R)
Thanks for your consideration.

[sebffischer]

Thanks a lot, this seems generally useful!

Please:

1. change the function name to something shorter that does also not use a . and camelCase.
2. Add a test
3. Update NEWS.md and document

[tdhock]

Thanks for the feedback @sebffischer is the name fun ok?

[sebffischer]

Thanks for the feedback @sebffischer is the name fun ok?

How about reduce_fun ?

[tdhock]

are you sure? you said it should be camelCase, and reduce_fun is not?

[sebffischer]

Sorry maybe my English was imprecise. It should neither be camelCase nor contain a .

[tdhock]

Hi I saw that docs are inherited from reduceResultsList, so I figured the easiest way to document the new argument would be to just give it the same name, fun, is that OK?

[sebffischer]

Yes, that makes sense!

[sebffischer]

Thanks again! :)