How do you decide when a black-box model is acceptable versus when an interpretable model is required?
