RF OOB error is not unbiased but at least not optimistically biased

slds-lmu / lecture_i2ml

I2ML lecture repository

https://slds-lmu.github.io/i2ml/

Creative Commons Attribution 4.0 International

142 stars 79 forks source link

RF OOB error is not unbiased but at least not optimistically biased #1212

Closed manuelhelmerichs closed 1 month ago

manuelhelmerichs commented 3 months ago

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 1 month ago

i dont think we claim anything wrong anymore? @manuelhelmerichs ?

manuelhelmerichs commented 1 month ago

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.