it is marked as maximise=True, meaning methods are rewarded for maximising RMSE. This seems like a bug.
It should probably be modified to be scale-invariant. If you multiply a DR by 10 it is no better or worse, but this metrics is penalizing methods that don't rescale their axes.
main
branch of openproblems.RMSE has two issues:
maximise=True
, meaning methods are rewarded for maximising RMSE. This seems like a bug.