search

mlr-org / mlrMBO

Toolbox for Bayesian Optimization and Model-Based Optimization in R
https://mlrmbo.mlr-org.com
Other
187 stars 47 forks source link

Check whether model (and settings), ctriteria, optimizer and so on work together. #14

Closed jakobbossek closed 10 years ago

jakobbossek commented 10 years ago

Should be done in checkStuff function

berndbischl commented 10 years ago

Check this whether it is reasobly good now. then close.

jakobbossek commented 10 years ago

We have a lot of checks in checkStuff now. Why is this stuff commented out?

  # for EI we need mu + sd
  #  if (control$infill.opt == "EI" &&
  #          !(class(learner) %in% c("regr.km", "regr.kmforrester")))
  #      stop("Expected improvement can currently only be used with learner 'regr.km' and 'regr.kmforrester'!")
  # if (!(inherits(learner, "regr.randomForest") || inherits(learner, "regr.rpart")) && hasRequires(par.set))
  #   stop("Parameter sets with dependent parameters currently require the learner to be a randomForest or rpart.")

Edit: I am going to remove the first uncommented check. EI works with all learner that support standard error estimation. This is already checked. The second one relates to dependent params. This is your job Bernd ;-)

jakobbossek commented 10 years ago

I think we should assert that the learner has the class RLearnerRegr?

jakobbossek commented 10 years ago

Forget it. We already check this :+1:

jakobbossek commented 10 years ago

Our simple EA optimizer can handle only numeric parameters as well as cmaes. We check this now.

jakobbossek commented 10 years ago

Closing this for now.