v0.5.2

v0.5.2

Breaking changes

• leidenalg is no longer installed by default. It is required by estimate_balancing_weight (called by fit_SCGLUE unless skip_balance=True) and normalized_mutual_info, and can be installed via pip install scglue[leiden].
• Renamed calc_hypo_score to calc_hypo_ratio.
• The "majority" aggregation in aggregate_obs and aggregate_var now returns NaN when the top category accounts for less than majority_cutoff (default 0.6) of a group. Set majority_cutoff=0 to restore the previous behavior.
• get_metacells now writes metacell assignments to obs["metacell"] of the input datasets.

New features

• Added support for AnnCollection in configure_dataset and model training.
• Added support for dask arrays in model training and aggregate_obs.
• Added a "join" aggregation method for discrete columns and a majority_cutoff option in aggregate_obs and aggregate_var.
• get_metacells now records the metacell each cell belongs to (Resolves #138).

Enhancements

• Moved configure_dataset to scglue.data. It remains accessible as scglue.models.configure_dataset.

Bug fixes

• Fixed compatibility with recent versions of anndata and pandas, including pandas 3 (Resolves #140).
• Fixed aggregate_obs failing when the by column contains missing values.
• Fixed "majority" aggregation failing when all values in a group are missing.

Full Changelog: v0.4.1...v0.5.2