How would you design a loss function for a model where false positives are ten times more costly than false negatives?
