Fix raw formatter loading for empty detections and large outputs

[mbsantiago]

Summary

This PR improves the raw output formatter for larger prediction sets and fixes round-trip handling for clips with no detections.

Changes include:

• avoid np.stack on empty class score / feature arrays when saving raw outputs
• add n_jobs support for loading raw prediction files in parallel
• add optional progress reporting while loading raw predictions
• speed up raw loading by avoiding per-detection xarray .sel(...) calls
• preserve recording metadata correctly when loading raw predictions
• add regression tests for recording metadata, empty detections, and multiprocessing load

Testing

• uv run pytest tests/test_data/test_predictions/test_raw.py

[Copilot]

Pull request overview

This PR updates the raw output formatter to correctly handle empty detections, improve raw prediction loading performance, and add optional parallel loading/progress reporting; it also adds regression tests for these behaviors.

Changes:

• Avoid stacking empty class_scores/features arrays when saving raw outputs.
• Add n_jobs-based parallel loading and optional progress reporting during raw loads.
• Optimize raw loading by avoiding per-detection xarray .sel(...) calls and improve recording metadata round-tripping.

Reviewed changes

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

File	Description
tests/test_data/test_predictions/test_raw.py	Adds regression tests covering recording metadata round-trip, empty detections, and multiprocessing load.
src/batdetect2/outputs/formats/raw.py	Implements empty-detection save fix, parallel/progress-enabled loading, and faster dataset-to-prediction conversion.

Suppressed comments (2)

src/batdetect2/outputs/formats/raw.py:95

• If show_progress=True and tqdm isn’t installed (see optional import above), calling tqdm(...) will fail with a TypeError. Raise a clear error when progress is requested but tqdm is unavailable.

        if self.show_progress:
            iterable = tqdm(files, total=len(files))

src/batdetect2/outputs/formats/raw.py:106

• Same as sequential load: when show_progress=True but tqdm isn’t installed, wrapping pool.imap(...) with tqdm(...) will fail. Add an explicit check and raise a helpful ImportError.

        with Pool(self.n_jobs) as pool:
            if not self.show_progress:
                return pool.map(self.load_single_file, files)

            return list(

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.