What kind of feature would you like to request?

Additional function parameters / changed functionality / changed defaults?

Please describe your wishes

@Intron7 and I set up a nice implementation using GPU here: https://github.com/scverse/rapids_singlecell/pull/179/files#diff-483d6f872ddf4abd63e32af66a8566e0dcb40ba853a8672771dcbffb0235b7f9

It should be straightforward (and easier) without GPU. Basic outline of computational-heavy steps:

1. Chunked covariance matrix calculation
2. Use already existing mean-var
3. Eigenvalue decomp

It works well on GPU so hopefully we can get nice performance on GPU.

I suspect using numba-in-dask for the covariance matrix calculation (similar to @Intron7 's kernel) would be super helpful.