Repository navigation
Convert: h5ad ↔ zarr #22
Description
Goal
adata convert data.h5ad -o data.zarr [--zarr-format 2|3]
adata convert data.zarr -o data.h5ad
This already works implicitly through adata subset in.h5ad -o out.zarr, which proves copy_store_contents handles every backend pair. A dedicated, discoverable command is what's missing.
Plan
- Thin converter over
storage.copy_store_contents. - Keep the source's chunking and compression by default; add
--rechunkand--compressorfor zarr targets. - Write consolidated metadata for zarr targets.
- zarr v3 by default from an h5ad source (existing behaviour); follow the source format for zarr → zarr.
- Refuse
-o -and output == input.
Tests
- All six pairings of h5ad / zarr2 / zarr3.
- Open the result with
anndata.read_zarr/read_h5adunder theintegrationmarker. - Perf guard: one read per chunk, no full materialisation.
Activity
adata convertto and from h5ad/zarr #20Claptar commentedon Oct 14, 2026
Implemented in #33, with one part deferred:
--rechunk. Chunking is carried over from the source, and the copy already sizes its reads by bytes (32 MiB) rather than by source chunks, so a badly chunked source is read efficiently anyway. Rewriting the chunk layout is a separate option that can follow when someone needs it.--compressionis in (none/gzip/lzffor h5ad,none/gzip/zstd/bloscfor zarr), and consolidated metadata is written for zarr outputs.