Oh, thanks for the correction. If all the vectors are on the unit ball, then cosine = dot product. But then the dot product is a linear transformation away fro

Oh, thanks for the correction. If all the vectors are on the unit ball, then cosine = dot product. But then the dot product is a linear transformation away from the euclidean distance: https://math.stackexchange.com/questions/1236465/euclidean-distance-and-dot-product https://math.stackexchange.com/questions/1236465/euclidean-d... If you're using it in a machine learning model, things that are one linear transform away are more or less the same (might need more parameters/layers/etc.) If you're using it for classical statistics uses (analytics), right, they're not equivalent and it would be good to remember this distinction.