Closed giuseppec closed 7 years ago
Is it sufficient to also upload the makeRLearner.regr.mycustomlearner
, trainLearner.regr.mycustomlearner
and predictLearner.regr.mycustomlearner
objects together with the learner object?
Would be great if this could be discussed soon, please.
I wanna test the new ctree: https://github.com/HeidiSeibold/sandbox/blob/master/rstuff/new_ctree_mlr.R
@HeidiSeibold you can still test this. Just create your own mlr fork that includes your custom learner and do something like to at least ensure that this is "reproducible" when people read:
flow = convertMlrLearnerToOMLFlow(makeLearner("classif.heidis.ctree"))
flow$description = "Please use the mlr version from LinkToHeidisMLRFork"
uploadOMLFlow(flow)
Or is something else blocking you?
I didn't want to spam OpenML with stuff that isn't standardized. I'll just try to get it to a reasonably reproducible level. To reproduce people need
makeRLearner.classif.newctree = function() {
makeRLearnerClassif( ...,
note = "Devel partykit package revision 1034: https://r-forge.r-project.org/scm/viewvc.php/pkg/devel/partykit/?root=partykit&pathrev=1034"
)
}
flow$description = "Please use the mlr add-on code https://github.com/HeidiSeibold/sandbox/blob/ed03326adacf4469f994b0c23ac4ecb0cb013ba3/rstuff/new_ctree_mlr.R"
Would that work? Should both links be in the flow$description
?
Looks good to me. You could also add both links to the flow description, they don't hurt anybody. And don't worry because of "spamming" you are doing it much more properly than you think.
I don't think we will support this with mlr. However, this might be possible with mlrng. Therefore closing for now.
Currently people can write their own mlr learners and create flows for custom mlr learners, however, since the
makeRLearner
,trainLearner
andpredictLearner
S3 function for custom flows are not uploaded on OpenML, other people can't reproduce runs created by custom flows. We need a way how we could handle this properly in OpenML, @berndbischl @jakobbossek any ideas/suggestions?