Closed Intron7 closed 5 months ago
ok so scanpy released today so I'll also release later today.
@Intron7, I have a question: why in _construct_neighbor_graph_rsc
, you use adata = adata[:, adata.var.highly_variable].copy()
instead of adata = adata[:, adata.var.highly_variable]
?
The copy() crates the subsetted matrix and it's not a view anymore. That works a lot better with a lot with the following functions
This adds support for rapids-singlecell for the
over clustering
#110 . The results might differ a bit since rsc uses a brute-force knn approach and the leiden clustering is also slightly different.The overclustering step is now for 200k cells 15x faster (3min vs 12s).
It's also set up in a way where when rsc fails to import it will default back to scanpy.
Please don't merge this yet since it uses the syntax that will be released with rapids-singlecell 0.10.0. This will come out shortly after scanpy 1.10.0