search

zachmayer / caretEnsemble

caret models all the way down :turtle:
Other
226 stars 75 forks source link

predict.caretEnsemble defaults type='prob' which errors for RF and GBM #171

Closed flylo closed 8 years ago

flylo commented 8 years ago

For tree models where finding the probability of a positive isn't possible, it's impossible to use the predict methods for ensembles, stacks, or lists. No matter what I do, I only get the following error:

Error in [.data.frame(out, , obsLevels, drop = FALSE) : undefined columns selected

I'm getting the same issue as this SO post. His model isn't working because random forests don't support predict.train(...type='prob')

jknowles commented 8 years ago

This is really hard to diagnose without a minimal reproducible example. As @zachmayer noted on the original SO post, please provide one so we can debug the error and make a precise patch. I think this actually has to do with some changes to the underlying caret API.

jknowles commented 8 years ago

Reported over in topepo/caret#274