good techniques in practice, but i was hoping for some kind of entropy-coding implementation of NumPy that would allocate bits used for a symbol based on its frequency. this article does bring that up as a possible project.
You might be interested in https://caterva.readthedocs.io/en/latest/ + https://github.com/Blosc/cat4py
It’s not NumPy-compatible, alas, but it’s an in-memory n-dimensional array that can use less memory and supports various operations you’d want from NumPy.
See https://www.youtube.com/watch?v=lP7A7lpzD18 for talk about it.