I am implementing a custom loss function where the expectation is to predict the labels evenly/uniformly. i.e., Given a set of labels say - [1,2,3,4,5], we need to maximize