Better performance

[alexfikl]

We will need to look into that at some point, since at the moment everything is implemented in the most naive way possible :cry:. Some options worth thinking about.

Use np.convolve or similar for the convolutions

This should be fairly simple to implement. Is scipy.spatial.convolve faster?

Implement some kernels in Cython

Need to profile and see what's worth reimplementing. Is Rust better here?

Use numba or similar to JIT functions

Does this need any effort on our side or can numba just JIT from user code? At the very least we'd need to test it and see that

[alexfikl]

Would need to set up some benchmarking to keep track of this!