Spaces:
Running
Running
Commit History
14 days ago
-
Apply uniform min128 dataset scope and compact Overall rows
Verified signatureUse the user-authorized 137-dataset scope (mean valid sequence length >=128), retain source archives, and gate FreshRetailNet-LT and Kaggle Web Traffic pending corrected full inputs. Remove redundant Overall row subtitles.86c9d33 -
Merge latest C++ results and clarify Overall timing coverage
Verified signatureUnify published C++ profiles by dataset and canonical method, selecting the latest authoritative campaign and preserving unsupported/N/A conclusions. Show admitted timed-subset coverage without borrowing old timing. Preserve immutable science payload and ranking formula. Independent source/data reviews and 756 UI regression assertions passed.799b74b -
Publish accepted corrected datasets and gzip operator evidence
Verified signatureAdd 858 independently accepted corrected-five dataset rows and the full 41090-row operator byte-effect union with 970 recovered rows. Operator CSV evidence is losslessly gzip encoded and verifies byte-for-byte after decompression. Preserve existing results and history. Fresh114 remains pending and excluded; unqualified timings stay unavailable.d84ab71
16 days ago
-
Publish verified 164-dataset native matrix and recovery results
Verified signaturePublish the complete 164 x 85 terminal matrix, 174 independently accepted recovery pairs, and receipt-bound timing admission. Preserve N/A and failure coverage, exact support-set ranking, execution-host and adapter distinctions. Exclude GLZA. The ongoing 328-pair recovery remains pending until joint independent acceptance. All 28 files match the reviewed v6 freeze receipt.7abb61f
18 days ago
-
Add linked trophies for THULab benchmark datasets
Verified signature5b51039 -
Complete medium_posts results and restore dataset timing averages
Verified signatureIntegrate the verified C++, Java, and Python medium_posts measurement into the existing leaderboard. Use equal-dataset timing averages when selected results contain aggregate-only timing evidence, while retaining equal-column averages for fully column-backed results. Refresh scores, ranks, compact data, and provenance.26386d8 -
Average compression times over measured columns
Verified signatureCompute Compress and Decompress ns/point independently for each measured column, then use an equal-column arithmetic mean across the selected datasets. Preserve compression rates, sizes, configurations, and dataset coverage. Keep aggregate-only historical timings unavailable. Update leaderboard scores, ranks, metric labels, charts, details, and CSV exports.43d6fa3 -
Align dataset summary with measured coverage
Verified signatureUse the same measured dataset selection for the summary count and coverage denominator. The default Overall view consistently shows 164 measured datasets. Keep all 169 metadata entries in the dataset browser and preserve the published measurements, scoring and rank rules.3b49d32 -
Allow zero weights in composite scores and average ranks
Verified signatureInclude all 66 non-negative one-decimal triples summing to 1.0, including the three single-metric endpoints. Keep the default 0.4/0.3/0.3 and the existing ranking and missing-measurement rules. Update weight validation and help text.8e46632 -
Integrate verified DBP measurements into the leaderboard
Verified signatureReplace float measurements with verified full-frame results, recompute available numeric aggregates, and use the shared score and rank rules. Retire the standalone measurement page and resources.93460be -
Remove method-specific Overall annotations
Verified signatureRemove the leaderboard notice, row badges, and method-detail audit annotations so all compression methods share the same presentation. Keep measured results, rankings, method documentation, and independent audit data.218d851 -
Publish lossless DBP native-float remeasurement
Verified signatureComplete 169-dataset native float audit: 162 measured datasets and 4096 nonempty float columns. All C++, Java and Python native values verified bit-exact. Encode and decode timing include scale/mode processing; frame bytes include dtype/mode/scale/count metadata. New column downloads and audit page remain separate from historical Overall ranks. Timing is a shared-server single-pass diagnostic with four workers.d52f140
19 days ago
-
Align final operator implementations and 169-dataset catalog
Verified signature3d62f04 -
Add method and operator descriptions with presentation references
Verified signatured70efe6 -
Remove Paper snapshot page and navigation
Verified signature0d2d110 -
Add weighted composite scores and average dataset ranks
Verified signatureAdd positive one-decimal a/b/c weights summing to 1.0, Overall composite scores, mean per-dataset fractional ranks, and reproducible CSV fields. Document DBP kernel timing and float reconstruction limits. Measurement records and aggregates are unchanged.fe81ad7
23 days ago
-
Publish compression benchmark results and interactive leaderboard
Verified signaturePublish the current 19-dataset result snapshot with separate C++, Java, and Python leaderboards, filters, dataset details, trade-off plots, and CSV export.bfe13e2