rudymartin/georsct
GeoRSCT A geospatial regression benchmark for evaluating representation–solver compatibility. GeoRSCT is a benchmark and evaluation framework for studying when geospatial model performance reflects solver quality versus target difficulty, spatial leakage, aggregation effects, scale sensitivity, or representation–solver mismatch. This release (version 24.0.1) includes 31,789 U.S. ZIP Code Tabulation Areas (ZCTAs), 106 columns spanning 33 ACS features, 37 geospatial enrichment… See the full description on the dataset page: https://huggingface.co/datasets/rudymartin/georsct.
Weekly refresh of live comparison matrix
Weekly refresh of live comparison matrix
chore(checksums): add georsct_simplified_001.geoparquet sha256 (v24 rebuild)
feat(v24): rebuild georsct_simplified_001.geoparquet from v24.002 data (106 cols + geometry)
chore: remove stale validation_report (2026-05-01)
fix(meta): refresh georsct_checksums.sha256 to v24.002 vintage
fix(meta): refresh georsct_schema.json to v24.002 vintage
docs(readme): fix all v24.0.1 inconsistencies — counts, lags, files table, citation, stubs
docs(readme): bump to v24.0.1; update feature counts, geometry snippet, NOAA sidecar example
chore(cleanup): remove stale geoparquet + old croissant variants; bump to v24.0.0
chore(manifest): update v24.002 column list and enrichment layers
fix(croissant): prune geoparquet ghost; expand cdc-ci + acs-moe stubs to full schemas; add noaa-storm-events-long sidecar recordSet (Version C Experiment 1)
feat(sidecar): add noaa_storm_events_long.parquet -- 979,939 rows x ZCTA x year 1996-2024 for Version C Experiment 1
chore(manifest): update v24.002 column list and enrichment layers
fix(croissant): reconcile schema to parquet — prune ghosts, add TWI descriptions, v24.002
chore(manifest): update v24.001 column list and enrichment layers
feat(v24.001): add NFIP + NOAA flood enrichment layers
chore(croissant): add NFIP + NOAA fields, bump to v24.001
chore(manifest): update v24.001 column list and enrichment layers
feat(v24.001): add NFIP + NOAA flood enrichment layers
chore(manifest): update v24.001 column list and enrichment layers
feat(v24.001): add NFIP + NOAA flood enrichment layers
chore(croissant): add NFIP + NOAA fields, bump to v24.001
Weekly refresh of live comparison matrix
Weekly refresh of live comparison matrix
Weekly refresh of live comparison matrix
fix: add contentSize to all FileObjects, remove self-referential entry
fix: add SHA256 hashes to all Croissant FileObjects for NeurIPS validation
fix: make Zenodo and SSRN citations into clickable links
fix: clarify ZCTA vs ZIP code relationship in field schema
fix: feature_columns() returns all 63 features, not just ACS
fix: remove built-environment from description fields
fix: remove built-environment from description fields
fix: remove built-environment from Croissant description fields
fix: align target families to 21/3/3 (Health/Socioeconomic/Environmental)
Upload folder using huggingface_hub
fix: standardize zcta_id key in README.md
fix: standardize zcta_id key in zcta_latents_v1.npz
fix: standardize zcta_id key in gnn_v2_latents.npz
add representation artifacts for experiment reproducibility (GNN latents, PCA, spatial lag transforms)
fix: specify HIFLD 2022 vintage in dataset card
feat: add population and flood_sfha fields to Croissant (107 fields)
fix: rename drive_min_to_county_centroid, update column counts
update: croissant.json (v24 - county_centroid rename, HIFLD 2022)
update: georsct_table.parquet (v24 rebuild)
update: georsct_simplified_001.geoparquet (v24 rebuild)
add: zcta_acs_margins_of_error.parquet uncertainty sidecar
add: cdc_places_ci.parquet uncertainty sidecar
fix: manifest version 23.0.2
fix: schema version 23.0.2