Datasets:
25 days ago
-
rel-f1: Ergast database images are CC BY-NC-SA 3.0
26 days ago
-
Label each database with its upstream license
-
Label per-database licenses; fix card metadata
-
Citations verbatim from source repos
-
Dataset-specific citations
-
Minimal card
-
Fix rel-stack task descriptions to match 91-day window
-
Flag known temporal leakage issues in task SQL (relbench#402)
-
rel-stack: regenerate post-votes, user-post-comment labels from the manifest SQL
-
rel-event/user-repeat: SQL that works for many timestamps at once; train labels regenerated from it (attended in the previous two windows)
-
Add regression-target stds (normalize MAE -> NMAE)
-
Point at the canonical website URL
-
Sort tables without a primary key by their time column
-
Add regression-target stds (normalize MAE -> NMAE)
-
Add STATS/tasks.parquet (tasks subset of the dataset viewer)
-
Add STATS/databases.parquet (databases subset of the viewer)
-
Update README for the v1/v2-extra split
-
Move v2-only tasks to stanford-star/relbench-v2-extra
-
manifests: plain block scalars for sql
-
STATS: task_type recommendation, metric map, tasks_recommendation column
-
task_type link_prediction -> recommendation; manifests as YAML block scalars
-
rel-event/user-repeat: kind forecast, not external -- it is an entity task like its siblings
-
Write task labels in canonical sorted order (row reorder only)
-
Write task labels in canonical sorted order (row reorder only)
-
move db-task-lists to the preprocessed repo
-
restore autocomplete train/val parquets
-
autocomplete tasks: drop hosted train/val labels (deterministically regenerable from the db; test.parquet kept for db-free leaderboard eval)
-
fix db-task-lists/all.json: curated relbench task kinds
-
fix db-task-lists/autocomplete.json: curated relbench task kinds
-
fix db-task-lists/forecast.json: curated relbench task kinds
-
add db-task-lists/all.json
-
add db-task-lists/autocomplete.json
-
add db-task-lists/forecast.json
-
STATS: size_gb = actual dataset-folder size
-
STATS: size_gb = whole dataset-folder size (db+tasks), fix #373 update
-
STATS: refresh size_gb after #373 column cleanup
-
rel-event: drop NaN/leakage/artifact columns (#373); refresh schema
-
rel-trial: drop NaN/leakage/artifact columns (#373); refresh schema
-
rel-stack: drop NaN/leakage/artifact columns (#373); refresh schema
-
rel-stack: remove multiclass badges-class task; drop empty multiclass column from STATS
-
Move overview tables into STATS/ (first folder in the Files tab)
-
tasks.parquet: rename dataset->database, drop inapplicable columns, reorder
-
Viewer: name the split 'eval'
-
Viewer: databases + tasks as separate configs; typed tasks.parquet
-
Viewer: rename overview->databases.parquet, add tasks.parquet split
-
Remove overview.csv (replaced by overview.parquet)
-
Add overview.parquet
-
Add regression-target stds (normalize MAE -> NMAE)
-
Remove orphan line left from overview.csv prose removal
-
Fix card frontmatter (real YAML) + viewer config; no overview prose link
