🎯 🐕

Racing backtest

Temporal-split evaluation of the greyhound ranker: train on races before the split date, freeze, and score the held-out later races out-of-sample on NDCG@1, top-1 hit-rate and the winner's negative log-likelihood. Each run executes as a background job (visible in Jobs) and lands in the leaderboard.

Blank split defaults to the median race off-date. Tick All range to span the whole warehouse.

Model runs leaderboard
TB_rank_ndcg_1783350136000 TB_rank_ndcg TB 0.2201 142633 13711 trained 2026-07-06 15:02
TB_rank_ndcg_1783344946000 TB_rank_ndcg TB 0.8958 1669 160 trained 2026-07-06 13:35
TB_rank_ndcg_1783343508000 TB_rank_ndcg TB 0.8472 1669 160 trained 2026-07-06 13:11
TB_rank_ndcg_1783341349000 TB_rank_ndcg TB 0.8333 1669 160 trained 2026-07-06 12:35
TB_rank_ndcg_1783339624000 TB_rank_ndcg TB 0.8264 1669 160 trained 2026-07-06 12:07
TB_rank_ndcg_1783338264000 TB_rank_ndcg TB 0.5903 1669 160 trained 2026-07-06 11:44
GH_rank_ndcg_4yr_1783271234000 GH_rank_ndcg_4yr GH 0.3478 161004 22994 trained 2026-07-05 17:07
GH_rank_ndcg_1783213783000 GH_rank_ndcg GH 0.5559 129771 18530 trained 2026-07-05 01:09
GH_rank_ndcg_1783184504000 GH_rank_ndcg GH 0.7194 20363 2559 trained 2026-07-04 17:01
GH_rank_ndcg_1782871790000 GH_rank_ndcg GH 0.9870 1842 230 trained 2026-07-01 02:09
GH_rank_ndcg_1782835364000 GH_rank_ndcg GH 0.6454 9916 1198 trained 2026-06-30 16:02
11 rows
Page 1 / 1