New in this version:
Fixed
get_rankings()no longer truncates average ranks to integers: tied genes
now keep fractional ranks (e.g. 812.5), as in the R version of UCell.- Fixed a bug in
get_rankings()where, for cells with a group of tied genes
spanningmax_rank, some genes were dropped from the ranking regardless of
their expression. Tied groups are now kept or dropped as a whole. - With these fixes, scores match the R version of UCell on full-transcriptome
data. Scores may change slightly compared to previous pyUCell versions.
Changed
- Ranks are stored as
float32instead ofint32on the CPU path (same
memory footprint). - Removed the
numpy<2pin; pyUCell now supports numpy 2 and Python 3.14. - Documentation build fixes for Read the Docs.