Changelog¶

Version 0.4.1¶

Added the aggregate_reference() function to aggregate references for speed. This is conveniently used via the aggregate= option in train_single().
Extend train_single() to support marker detection using scrapper to compute Cohen’s d or the AUC of each pairwise comparison.

Switch to using pybind11 for the Python/C++ interface.
Update to the latest singlepp C++ library.
Store marker lists as lists of StringList objects in the metadata of the BiocFrame objects.
Automatically remove None labels in the reference.
Automatically remove duplicated feature names in the reference.
Remove the ref_names= option in train_integrated() for simplicity.
Remove support for auto-loading of references based on their names; users should extract references from celldex instead.
Deprecate options to extract features/labels from a SummarizedExperiment based on the column name; users should explicitly pass the column contents to each function.

Compatibility with NumPy 2.0

Migrate references to the celldex package.

Added the annotate_integrated() function, which performs annotation across multiple references.