Closed Intron7 closed 1 year ago
I was just able to confirm that this version is a lot faster than the version in the newest release. On my A100 80GB It runs in 1m 4 sec vs 1m 51 sec. It also avoids some errors that are still present and in the publication.
@Intron7, I know it's been awhile since you've opened this PR. If you can fix the merge conflicts here then I'll give it a review.
Dear all,
I fixed the
rank_genes_groups
function. However I change quit a bit of the preprocessing e.g. masking from cupy to numpy. In the notebooks that I was able to test (so not the 1Million cell notebooks) this didn't negatively impact performance. I might have even sped it up a bit. However I don't know if this will be the case even for the 1 Million cells notebooks. So you might want to test this first before merging. I also changed the the input so that it now uses theadata
object with agroupby
variable that takes an.obs
column. If you want me to I can also include GPU functions for diffusion maps and draw_graph (force atlas 2).