Closed
Description
There are situations when having vectors y_pred and y_true is not sufficient to compute the score/loss function. An easy example would be a mean (squared / absolute) error weighted by a pre-defined column of weights. In this case, this column vector has to be sliced by the cross-validator the same way X and y are sliced, and the sliced vector is to be passed to the score/loss function. While I understand that GridSearchCV can be (fairly painlessly) modified to account for this (say, an optional parameter to the score/loss func), I was wondering if anyone else ever faced the same problem.