We probably need to add a custom loss function for regression problems rather than "copy paste" for logistic