Datasets:
Update bioclip-2_float16: metadata backfill + UUID dedup
#2
by netzhang - opened
Changes
Metadata backfill
- 27.2M observation.org rows recovered: taxonomy from resolved_taxa (2025-12-20), URLs from GBIF data parquets
- 1.1M EOL unresolved rows recovered: source_dataset and source_id from EOL metadata
- All taxonomy refreshed against latest resolved_taxa (2025-12-20, all 4 sources)
- 0 NULL source_dataset remaining (was 28.3M)
UUID fixes
- 27.2M non-hyphenated UUIDs normalized to hyphenated format
- 21,078 duplicate UUIDs removed (exact duplicates from embedding pipeline; source images have 0 dups)
Stats
- Rows: 239,580,103 → 239,559,025 (after dedup)
- Size: 331 GB → 332 GB
- Kingdom coverage: 88% → 97%
- Species coverage: 81% → 90%
- Validation: 9/9 checks passed
netzhang changed pull request status to closed
netzhang deleted the
refs/pr/2 ref