Answer to: Optimizing a Gaussian penalty function for HSL color compatibility in PyTorch/NumPy

I don't know about the details of this color harmony system, but the usual way to compare angular values is with trigonometric functions. sin(0) = 0, sin(90) = 1, cos(0) = 1, etc... TensorFlow probably has an efficient implementation of those. On the other hand, the implementation you have may well be more efficient, you would probably have to benchmark them both to know which is better.