-
> Make dataset viewer‑ready: preview/dev Parquet, scripts, README, .gitignore
- Add tiny preview/dev manifests with shapes, paths, and checksums:
- manifests/preview.jsonl (embeds 8×8 correlation_matrix), manifests/dev.jsonl (metadata-only)
- Parquet mirrors: manifests/preview.parquet, manifests/dev.parquet
- Add precomputed 8×8 tiles for preview: preview/<parc>__<subject>__corr8x8.json
- Add helper scripts:
- scripts/enrich_manifests.py (shapes from sidecars, checksums from manifest.jsonl, embed 8×8 for
preview)
- scripts/jsonl_to_parquet.py (JSONL → Parquet)
- scripts/scan_to_manifest.py (scan data/ to build metadata-only manifests)
- scripts/make_preview.py (generate 8×8 tiles from .mat using SciPy/mat73)
- Add parcellations.json (parcellation → ROI count)
- Update README:
- YAML tags for discoverability
- Modern datasets usage via data_files (no Python dataset script)
- Preview vs dev behavior, variable fallbacks (correlation_matrix|corr|A; features_timeseries|
timeseries|X)
- Integrity/…
