Ranger does not throw an error when predicting with a categorical variable with new/unseen categories

[D-Maar]

Normally an algorithm (e.g. lm) throws an error, if one tries to predict with unseen categories in a categorical variable. ranger just predicts something....
Code:

a<-data.frame(a1=c(1,2,3,4), a2=c("d", "a", "d", "e"))
res<-ranger::ranger(a1~a2,a)
predict(res, data = data.frame(a1=c(1,2,3,4), a2=c("d", "a", "f", "f")))$prediction

[mnwright]

I agree, this should give an error (or maybe a warning and document what happens with the new category).