Preserve arbitrary pose HDF5 attributes in NWB conversion

[gbeane]

Summary

When converting a JABS pose HDF5 file to NWB, any attributes stored in the source file that JABS does not explicitly parse were previously dropped. This change captures every attribute from the entire pose HDF5 file and carries it into the NWB output as metadata, so arbitrary provenance attributes are not lost.

What changed

src/jabs/scripts/cli/convert_to_nwb.py:

• _collect_hdf5_attributes(path) walks the whole file (root group, all sub-groups, all datasets) via h5py.visititems and records each object's attributes keyed by its HDF5 path ("/", "poseest", "poseest/points", …). Objects with no attributes are omitted.
• _h5_attr_to_jsonable(value) normalizes h5py return types (numpy scalars/arrays, bytes from fixed-length string attrs) into plain JSON-friendly types. An unrecognized type is preserved as str(value) with a warning rather than dropped.
• pose_to_pose_data stores the result under metadata["hdf5_attributes"].

No NWB-adapter changes were needed: PoseData.metadata already flows through the writer into the jabs_metadata scratch JSON and is recovered on read, giving a lossless round-trip.

Design note

All attributes are captured verbatim, including ones JABS already parses separately (version, cm_per_pixel). This preserves the raw on-disk values exactly and avoids a fragile allowlist that would need maintenance as pose formats evolve.

Testing

• New unit tests cover scalar/array/bytes normalization, the unsupported-type fallback, full-file collection (incl. empty-group omission), and JSON-serializability.
• All 52 existing NWB adapter tests pass (no regression).
• A real PoseData → NWB → PoseData round-trip confirms a nested hdf5_attributes dict comes back identical.
• ruff check and ruff format clean.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR updates the NWB conversion pipeline to preserve all HDF5 attributes from JABS pose files by collecting them across the entire HDF5 object tree and embedding them into PoseData.metadata["hdf5_attributes"], with normalization to JSON-serializable types.

Changes:

• Added _collect_hdf5_attributes(path) to traverse an HDF5 file (root/group/dataset) and collect attributes keyed by HDF5 object path.
• Added _h5_attr_to_jsonable(value) to normalize h5py/numpy attribute values into JSON-friendly Python types (with a warning-based string fallback).
• Added unit tests covering scalar/array/bytes normalization and full-file attribute collection + JSON serializability.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/jabs/scripts/cli/convert_to_nwb.py	Collects and serializes all pose-file HDF5 attributes into PoseData.metadata during NWB conversion.
tests/scripts/test_convert_to_nwb.py	Adds tests for attribute normalization and attribute collection behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.