🎯 🏇
← Racing backtest

GH_rank_ndcg trained

GH_rank_ndcg_1782871790000 · GH · rank:ndcg · 2026-06-20 → 2026-06-30

0.9870
NDCG@1
0.9870
Top-1 hit
0.9000
Race NLL
230
Races
1842
Rows
43
Features

Feature importance (gain)

Mean split gain per feature from the persisted booster (top 15). ELO-family features are expected to dominate the greyhound ranker.

Feature columns

age_days age_months best_time_track_distance_pit career_start_count career_win_rate_pit career_wins days_since_last_start days_since_prev distance_change_m dog_days_since_last_win dog_elo_avg_mov_surprise dog_elo_mov_delta dog_elo_percentile_in_field dog_elo_rating dog_top3_rate_30d dog_win_rate_30d elo_diff_vs_rival_med elo_maturity elo_vs_field_avg field_diff_best_time_track_distance_pit field_diff_career_win_rate_pit field_diff_place_inv_mean_365d field_diff_prize_money_career_sum field_rank_best_time_track_distance_pit field_ratio_best_time_track_distance_pit field_ratio_career_win_rate_pit field_ratio_place_inv_mean_365d field_ratio_prize_money_career_sum field_size grade_best_ever grade_ordinal place_inv_mean_365d place_inv_mean_91d place_log_mean_365d prize_money_career_sum race_field_elo_mean race_field_elo_std rival_elo_mean rival_elo_median trainer_elo_maturity trainer_elo_rating weight_delta_3 weight_last

Hyperparameters

colsample_bytree
0.7
eta
0.05
gamma
0.0
lambdarank_num_pair_per_sample
8
lambdarank_pair_method
"topk"
max_depth
6
min_child_weight
1
ndcg_exp_gain
false
num_boost_rounds
200
objective
"rank_ndcg"
reg_alpha
1.0
reg_lambda
5.0
seed
42
subsample
0.8
tree_method
"hist"

Predictions (0 runners across 0 prediction runs)

No predictions produced by this model yet.
0 rows
Page 1 / 1