Notebook Gallery¶

These notebooks demonstrate BADC workflows using the public bogus dataset. Keep them lightweight (≤10 MB inputs, stub inference by default) so contributors can run them on laptops or CI workers without GPUs.

Layout overview¶

chunk_probe.ipynb: Explore badc chunk probe / badc chunk split and visualize segment plans.
infer_local.ipynb: Run badc infer run in stub mode, inspect dataset-aware outputs, and aggregate results.
aggregate_analysis.ipynb: Load detection JSON/CSV/Parquet plus badc report quicklook CSV exports into pandas/DuckDB, then generate quick sanity plots (top labels, chunk timelines).

Execution guidelines¶

Use badc data connect bogus so manifests/audio stay inside data/datalad/bogus.
Start each notebook with reproducible bootstrap cells (pip install -e ., env vars, etc.).
Mark GPU-heavy cells with USE_HAWKEARS guards or convert them into badc --print-datalad-run snippets until we have GPU-backed CI.
Clear outputs before committing; if we need rendered previews, export HTML copies into docs/.