昨日からちょっとまたガッツリとコーディングしてたので予想の時間とか全く取れてない。でもまあ、GIなので参加しなきゃって事で^^; 今回はたった今終わった相対タイムの24時間学習にマル乗りで行きます。\(R^{2}\)値は0.2746です。
随分と堅い予想ですが^^;
まあ、軽く乗ってます(笑)
で、そのモデルの検証ですが、
| 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 | 24.75% (77.99%) |
24.57% (74.42%) |
24.85% (81.08%) |
25.83% (85.67%) |
37.90% (88.40%) |
26.45% (76.87%) |
22.99% (77.42%) |
53.95% (75.42%) |
| 複勝 | 56.03% (83.79%) |
56.07% (82.63%) |
55.77% (84.60%) |
59.17% (89.08%) |
69.41% (90.73%) |
60.32% (81.30%) |
53.40% (84.00%) |
87.38% (81.44%) |
| 枠連 | 14.21% (79.98%) |
13.91% (79.38%) |
13.98% (78.06%) |
22.22% (119.09%) |
-- (--) |
15.70% (69.69%) |
12.81% (78.34%) |
28.59% (78.06%) |
| 馬連 | 11.23% (70.86%) |
11.60% (75.87%) |
10.69% (66.00%) |
13.33% (67.00%) |
26.48% (101.74%) |
12.11% (58.83%) |
9.56% (72.08%) |
23.36% (69.92%) |
| ワイド | 24.66% (78.37%) |
25.22% (76.59%) |
24.06% (81.29%) |
25.00% (63.33%) |
51.60% (92.47%) |
28.43% (75.03%) |
20.97% (78.21%) |
45.76% (79.24%) |
| 馬単 | 5.88% (64.18%) |
6.04% (65.03%) |
5.77% (64.64%) |
5.00% (45.83%) |
15.07% (85.84%) |
6.06% (54.81%) |
4.99% (65.34%) |
23.36% (68.12%) |
| 三連複 | 6.45% (84.95%) |
7.52% (77.78%) |
5.47% (95.95%) |
5.00% (34.92%) |
20.55% (102.92%) |
8.16% (74.88%) |
4.61% (86.68%) |
15.60% (70.57%) |
| 三連単 | 1.51% (56.48%) |
1.95% (51.64%) |
1.15% (65.57%) |
0.00% (0.00%) |
6.39% (115.80%) |
1.48% (38.54%) |
1.07% (57.11%) |
15.60% (64.34%) |
| 総合 | 57.19% (74.50%) |
57.25% (72.78%) |
56.80% (77.14%) |
61.67% (61.86%) |
73.52% (96.84%) |
61.56% (66.24%) |
54.26% (74.90%) |
87.64% (69.04%) |
まあ、まだまだ精度上げなきゃならないですが、1点買いの総合で74.5%の回収率は普通? 控除率的に買い続けるとこの辺りに収まるんでしたっけ? ちょっと検証方法の工夫も今後必要になるとは思ってるんですが、走破タイムの時にはあえて新馬戦とその他を分けました。相対タイムの学習では全てのレースを対象にしてますので、検証も全てのレースです。新馬戦込みでの検証なので、ちょっと新馬戦とそれ以外の検証も気になるんですよね。
ただ、今やっている作業はちょっと反れてます(笑)
追記 2026.5.17 22:02
一旦コーディングは休憩して、まあ、今度はAutoMLで6時間×10回スタート。まず初回は\(R^{2}\)値が0.2520で
| 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 | 25.18% (75.15%) |
23.92% (69.37%) |
26.37% (80.95%) |
26.67% (76.83%) |
36.53% (83.65%) |
26.95% (74.68%) |
23.57% (74.54%) |
56.15% (77.15%) |
| 複勝 | 56.82% (82.65%) |
56.36% (82.42%) |
56.99% (82.96%) |
60.83% (81.75%) |
69.41% (86.89%) |
61.80% (83.67%) |
54.02% (81.93%) |
89.23% (81.34%) |
| 枠連 | 14.28% (79.06%) |
13.98% (79.98%) |
14.17% (78.07%) |
20.20% (81.72%) |
-- (--) |
16.93% (72.58%) |
12.48% (76.20%) |
30.14% (78.80%) |
| 馬連 | 11.49% (75.33%) |
11.25% (69.53%) |
11.36% (81.16%) |
16.67% (77.00%) |
26.48% (84.20%) |
14.59% (78.01%) |
9.11% (73.63%) |
24.80% (73.52%) |
| ワイド | 25.33% (77.81%) |
25.52% (74.35%) |
24.97% (81.56%) |
27.50% (75.17%) |
52.51% (96.48%) |
31.77% (81.58%) |
20.73% (74.87%) |
47.06% (74.62%) |
| 馬単 | 6.19% (75.21%) |
5.62% (65.50%) |
6.56% (85.42%) |
9.17% (71.83%) |
15.07% (75.16%) |
7.29% (73.79%) |
5.03% (75.69%) |
24.80% (71.63%) |
| 三連複 | 6.19% (66.12%) |
7.16% (62.79%) |
5.29% (71.14%) |
5.00% (44.00%) |
19.63% (89.22%) |
8.16% (59.04%) |
4.33% (66.39%) |
16.24% (66.99%) |
| 三連単 | 1.30% (56.07%) |
1.66% (55.06%) |
0.97% (59.70%) |
0.83% (20.50%) |
3.65% (43.52%) |
1.98% (58.13%) |
0.87% (56.51%) |
16.24% (65.29%) |
| 総合 | 58.00% (73.35%) |
57.73% (69.66%) |
58.02% (77.62%) |
61.67% (65.75%) |
74.43% (79.88%) |
62.30% (72.69%) |
55.09% (72.47%) |
89.49% (69.73%) |
と、相変わらずModel Builderよりも\(R^{2}\)値は低いものの的中率は高い感じなんですよね。ただ、多分三連系が若干劣るのかなぁ。これ、12時間で行くか迷ったんだけど、まあ、小刻みにまずは回して当たりが出るか確認です。
