0.3.0 🌈 Release

Release adjusting for the anndata 0.13.0 release

Added

• {func}~ehrdata.io.read_h5ed and {func}~ehrdata.io.write_h5ed are the new primary HDF5 I/O functions: .h5ed is now ehrdata's on-disk format, marking it distinct from anndata's .h5ad. read_h5ad and write_h5ad remain as deprecated aliases. (#252) @eroell
• {func}~ehrdata.io.write_h5ed and {func}~ehrdata.io.write_zarr now write the ehrdata on-disk encoding 0.2.0. Because AnnData only guarantees 2D but not 3D arrays in X/layers (see scverse/anndata#2430), 3D arrays are relocated into .obsm (under reserved _ed_ondisk_X / _ed_ondisk_layers_<name> keys) and dropped from X/layers (#252) @eroell

Maintenance

• ehrdata is now compatible with anndata 0.13 (unified X/layers storage, IndexManager view indices, index type aliases moved to anndata.typing, and the stricter 2D-only X/layers on-disk rule) while remaining compatible with anndata <0.13. CI gains an integration-free core-anndata-min lane (pinned to anndata<0.13) and a pre-release lane to catch upstream breakage early. (#252) @eroell
• DaskArray import fixed to follow anndata 0.12.16, and dask added to the intersphinx mapping. (#248) @sueoglu
• CI now caches downloaded datasets used by ehrdata.dt to reduce flaky upstream hosts (e.g. physionet.org) breaking the test and notebook workflows. (#250) @eroell
• Dataset downloads now use pooch instead of a custom requests-based downloader, aligning with the scverse ecosystem and providing caching out of the box. (#251) @eroell
• The tqdm dependency has been removed. (#254) @eroell