自分のネタ的には既に底ついてる気もするので、AIを頼りにさらなる精度を求めて次の特徴量として系統を追加。AutoMLとModel Builderで学習を進めるんですが、AutoMLはまず1時間テスト学習して、Model Builderは2時間テスト学習。そのModel Builderの2時間で\(R^{2}\)値が0.2599と時間的には良い感じだったので検証すると
| 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 | 24.40% (76.27%) |
25.22% (76.76%) |
23.63% (74.59%) |
23.33% (92.50%) |
37.90% (84.34%) |
27.32% (80.43%) |
22.21% (74.16%) |
53.75% (79.06%) |
| 複勝 | 55.46% (82.89%) |
56.48% (83.13%) |
54.07% (81.70%) |
60.00% (95.75%) |
70.78% (91.28%) |
60.69% (82.62%) |
52.33% (82.22%) |
87.90% (84.00%) |
| 枠連 | 13.08% (71.52%) |
13.34% (76.59%) |
12.52% (66.10%) |
18.18% (85.66%) |
-- (--) |
14.83% (70.28%) |
11.66% (67.40%) |
28.03% (74.01%) |
| 馬連 | 10.59% (68.07%) |
11.37% (75.86%) |
9.66% (57.57%) |
12.50% (102.42%) |
26.48% (100.18%) |
12.61% (78.81%) |
8.49% (61.59%) |
23.01% (71.67%) |
| ワイド | 23.76% (77.24%) |
24.75% (78.44%) |
22.54% (74.51%) |
26.67% (97.67%) |
47.95% (86.03%) |
28.80% (78.78%) |
19.90% (75.93%) |
45.82% (83.06%) |
| 馬単 | 5.24% (55.46%) |
5.62% (56.70%) |
4.80% (49.62%) |
5.83% (118.17%) |
15.07% (90.32%) |
6.43% (78.95%) |
3.96% (44.49%) |
23.01% (68.69%) |
| 三連複 | 5.99% (77.14%) |
7.40% (88.80%) |
4.43% (67.37%) |
7.50% (47.00%) |
18.72% (79.77%) |
7.79% (65.20%) |
4.24% (80.88%) |
15.72% (81.14%) |
| 三連単 | 1.19% (71.55%) |
1.54% (107.19%) |
0.85% (37.09%) |
0.83% (42.67%) |
5.48% (85.66%) |
1.24% (50.89%) |
0.78% (77.17%) |
15.72% (75.43%) |
| 総合 | 56.56% (72.53%) |
57.55% (80.52%) |
55.10% (63.56%) |
62.50% (85.22%) |
75.80% (88.23%) |
61.31% (73.25%) |
53.23% (70.48%) |
88.08% (75.99%) |
期待以上に良い感じ。今はAutoMLの6時間してて、その後にModel Builderの24時間と行く予定なんですが、2つ同時にやるとどうしても弊害が気になるので試せない。その昔ならPCを数台とか持ってた事もありましたが...地道に進めます。
0 件のコメント:
コメントを投稿