randomForest hangs when nodesize >= sampsize

[mb706]

If the nodesize parameter (either given, or default 5 for regression and 1 for classification) exceeds (edit: or equals) the sampsize parameter (either given, or defaulting to if (replace) nrow(x) else ceiling(0.632 * nrow(x))), classif.randomForest and regr.randomForest hang and can no longer be interrupted with Ctrl-C.

> resample(makeLearner("regr.randomForest", sampsize=2, nodesize=3), pid.task, hout)
[Resample] holdout iter 1:    

I don't know if this should be mlr's responsibility; in principle, the trainLearner code could check for this and throw an error. Please tell me if these kind of bugs don't belong here.

[larskotthoff]

Hmm, we should be able to do this when #1126 is merged I think. And yes, certainly a good thing to raise here.

[berndbischl]

. Please tell me if these kind of bugs don't belong here.

i really do not think that we want to maintain all of these special cases in mlr?
and if the learner hangs that seems like a really bad bug in the package - which you should take up with the original package author

[larskotthoff]

But in this case it makes sense to check this in mlr. I agree that the learner hanging is a bug, but even if it didn't it wouldn't make sense.

[jakob-r]

But then still - in the very end randomForest should throw a useful error message.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.