RF OOB error is not unbiased but at least not optimistically biased
Closed this issue · 2 comments
manuelhelmerichs commented
In Learning Goals v. slides-forests-intro:
Explain this better: OOB error is not unbiased but (at least) not optimistically biased bc we do not violate the untouched test set principle. (von OL, heißt wenn resolved dort auch!)
berndbischl commented
i dont think we claim anything wrong anymore? @manuelhelmerichs ?
manuelhelmerichs commented
Yes, this has been corrected in the new forests-oob chapter ("As we do not violate the untouched test set principle, GE is not optimistically biased."), in the vid this is explained as well.