Changelog

Changelog#

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.0.0] - 2026-06-07#

Changed#

  • Migrated to scvi-tools ≥1.4 (Lightning 2.x): replaced use_gpu with accelerator/devices, updated SaveBestState → SaveCheckpoint, _initialize_model, and _get_loaded_data call sites.

  • GroupDataSplitter now splits at the sample level (not cell level) using external_indexing, ensuring every validation bag is complete.

  • Bumped requires-python to >=3.12; switched package manager to uv with a committed uv.lock.

  • numpy ≥2.0 and anndata ≥0.12 compatibility fixes (dtype identity check, pandas groupby(observed=True)).

Fixed#

  • Single-layer regressor head used z_dim instead of class_input_dim when sample covariates are present.

  • .squeeze() on 1-element tensors in create_df produced 0-d scalars; wrapped with np.atleast_1d.

  • Fixed floating-point undercount in GroupDataSplitter (floor(1.0 - 0.9) = 0 edge case).

  • Corrected misleading intro text in the regression tutorial notebook (incorrectly stated IPF samples were included; they lack age annotations and are excluded).

[0.3.2] - 2025-12-16#

Added#

  • tutorial for regression

  • support for sample-covariate embeddings, experimental, only one-hot

Changed#

  • removed muon from dependencies for tutorials

  • changed the classification tutorial to include training across 3 CV folds

Fixed#

  • fixed a bug in regression which was caused by a wrong key when accessing the continuous covariates registered with scvi-tools

[0.3.1] - 2025-07-14#

Fixed#

  • Fixed a bug in score_top_cells that didn’t set the specified key_added column in .obs to True.

  • Fixed a bug in score_top_cells that set the key_added to be categorical, which resulted in wrong indexing in get_sample_representations.

[0.3.0] - 2025-07-13#

Added#

  • Utility functions: Added score_top_cells and get_sample_representations to utils module

  • score_top_cells: Function to identify and score top cells based on attention weights

  • get_sample_representations: Function to aggregate cell-level data to sample-level representations

Changed#

  • Major refactoring: Removed MultiVAE and MultiVAE_MIL models, keeping only MIL classifier

  • Code cleanup: Removed MLP attention weight learning from Aggregator class

  • Parameter consistency: Fixed default values between model and module classes

  • Dynamic z_dim: Automatically infer z_dim from input data shape instead of hardcoded value

Fixed#

  • Make categorical covariates categorical: Ensure the correct type in setup_anndata.

  • Improved error handling: If the prediction covariate hasn’t been registered with setup_anndata, throw an error.

  • Dead links to API and changelog: Fixed in README.