ランキング学習に追加した特徴量が悪影響し単勝的中率が2～3%落ちてたので、その特徴量を無視する様にしたものです。
イテレーション数3,000、葉271、率0.01、分割数47、葉の最小50、ボーダー500、乗数2.0で学習させNDCG@1=0.7795 NDCG@3=0.7832というモデル
| 3,455R | 1点 | 芝(1,689R) | ダート(1,646R) | 障害(120R) | 8頭以下(219R) | 9～12頭(809R) | 13頭以上(2,427R) | 多点 |
| 単勝 | 28.45% (84.64%) |
27.41% (84.03%) |
29.22% (83.66%) |
32.50% (106.67%) |
35.62% (78.77%) |
29.91% (86.66%) |
27.32% (84.50%) |
58.35% (77.63%) |
| 複勝 | 59.68% (85.75%) |
59.21% (85.41%) |
59.90% (85.83%) |
63.33% (89.42%) |
70.32% (87.40%) |
64.40% (85.67%) |
57.15% (85.62%) |
89.52% (82.57%) |
| 枠連 | 16.00% (82.11%) |
16.47% (90.87%) |
15.25% (73.94%) |
21.21% (87.27%) |
-- (--) |
17.18% (75.74%) |
14.59% (79.02%) |
31.21% (77.55%) |
| 馬連 | 12.65% (81.45%) |
13.03% (83.90%) |
11.97% (79.78%) |
16.67% (69.75%) |
23.74% (61.42%) |
14.71% (78.68%) |
10.96% (84.18%) |
25.85% (73.79%) |
| ワイド | 27.15% (87.40%) |
28.42% (89.50%) |
25.70% (86.00%) |
29.17% (76.92%) |
48.40% (78.31%) |
32.51% (84.26%) |
23.44% (89.26%) |
49.70% (80.68%) |
| 馬単 | 7.18% (77.90%) |
6.87% (74.82%) |
7.23% (81.28%) |
10.83% (74.92%) |
13.24% (59.59%) |
8.16% (74.00%) |
6.30% (80.85%) |
25.85% (70.97%) |
| 三連複 | 6.51% (78.05%) |
7.40% (58.88%) |
5.35% (95.43%) |
10.00% (109.58%) |
18.26% (64.84%) |
8.90% (73.28%) |
4.66% (80.83%) |
17.19% (81.15%) |
| 三連単 | 1.62% (71.76%) |
1.84% (53.82%) |
1.40% (90.14%) |
1.67% (72.08%) |
4.57% (47.40%) |
2.47% (66.14%) |
1.07% (75.83%) |
17.19% (73.70%) |
| 総合 | 60.78% (81.12%) |
60.57% (77.37%) |
60.81% (84.56%) |
63.33% (85.79%) |
74.43% (68.25%) |
65.02% (78.05%) |
58.14% (82.51%) |
89.84% (75.40%) |
どれにするか悩みましたが、これを今回添付してます。
0 件のコメント:
コメントを投稿