infill.crit="ei" does not work with any model about mlrmbo HOT 9 CLOSED

It works with any regr. model that is able to provide predict.type="se". If the model does not, you still have the option to use makeBaggingWrapper to bootstrap the se estimator. Should be mentioned in tutorial, help text and maybe error message.

from mlrmbo.

Jakob, please for now check (in code) that the model can predict "se" for the infill crits that need "se".
Otherwise produce a helpful error message. Maybe mention the BaggingWrapper BRIEFLY as a solution, too.

Then close this

from mlrmbo.

checkStuff now checks, if the infill criterion needs se estimation and if the learner has support for that. If not the an error message is presented to the user with the hint to use makeBaggingWrapper.

from mlrmbo.

Huh?

We already had that check!

    if(control$infill.crit %in% c("ei", "aei") && !learner$se)
      stopf("The infill criterion needs the learner to support the prediction of standard error, but the provided learner does not (you could use the mlr wrapper makeBaggingWrapper to bootstrap the standard error estimator).")
    if(control$infill.crit %in% c("ei", "aei", "lcb") && learner$predict.type != "se")
      stopf("For infill criterion '%s' predict.type of learner %s must be set to 'se'!%s",
        control$infill.crit, learner$id,
        ifelse(learner$se, "",
          "\nBut this learner does not seem to support prediction of standard errors!"))

Why did you basically add the same lines (without lcb, which also need se)?

from mlrmbo.

Nadja, can you please explain you original problem?

from mlrmbo.

Also check that we have a unit test for EI with model != km!

from mlrmbo.

Yes, it works. At the moment I am using "ei" with randomForest for example.

from mlrmbo.

Nadja, do you have any remaining issues?

Other, Jakob, pls check the unit tests and then close this, when the checkStuuf method is fixed again

from mlrmbo.

Fixed checkStuff. We have a unit test for EI with model != Kriging already in test_infillcrits.

from mlrmbo.