SaraD_x64なら使えるだろうと
NumberOfLeaves: 399
NumberOfIterations: 3000
LearningRate: 0.01
MaximumBinCountPerFeature: 47
MinimumExampleCountPerLeaf: 80
FeatureFraction: 0.75
SubsampleFraction: 0.8
CustomGains: 0～500 = 0点, 501～1000は1.8乗に0～1000点
昨夜就寝前に50回スタートさせてました。流石に26MB程度のモデルになりました。しかし、最後の10回は今回で言うと全て無駄でした^^; まあ、そもそも一番良さそうなのは３回目のNDCG@1=0.7809 NDCG@3=0.7843
| 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 | 28.16% (81.38%) |
27.89% (83.51%) |
28.25% (78.57%) |
30.83% (89.92%) |
38.36% (84.89%) |
30.28% (87.79%) |
26.53% (78.93%) |
58.90% (78.56%) |
| 複勝 | 58.55% (83.50%) |
58.61% (84.45%) |
58.02% (82.44%) |
65.00% (84.75%) |
71.23% (87.90%) |
64.77% (86.51%) |
55.34% (82.10%) |
90.10% (83.39%) |
| 枠連 | 15.41% (78.26%) |
15.05% (83.09%) |
15.37% (73.01%) |
21.21% (93.03%) |
-- (--) |
16.93% (74.09%) |
13.93% (74.68%) |
31.93% (78.97%) |
| 馬連 | 12.19% (77.60%) |
12.26% (76.38%) |
11.85% (79.25%) |
15.83% (72.17%) |
24.20% (63.84%) |
14.22% (74.86%) |
10.42% (79.76%) |
26.14% (74.51%) |
| ワイド | 26.54% (82.78%) |
27.77% (82.62%) |
25.03% (83.18%) |
30.00% (79.50%) |
50.23% (83.97%) |
30.90% (80.26%) |
22.95% (83.51%) |
49.58% (81.03%) |
| 馬単 | 6.63% (77.20%) |
6.22% (73.61%) |
6.93% (80.52%) |
8.33% (82.33%) |
12.33% (56.35%) |
7.66% (83.99%) |
5.77% (76.82%) |
26.14% (71.18%) |
| 三連複 | 6.51% (75.86%) |
7.76% (67.60%) |
5.04% (82.25%) |
9.17% (104.50%) |
17.81% (61.87%) |
10.01% (84.88%) |
4.33% (74.12%) |
17.11% (74.01%) |
| 三連単 | 1.42% (56.58%) |
1.54% (62.31%) |
1.34% (40.19%) |
0.83% (200.75%) |
3.65% (47.08%) |
2.84% (86.21%) |
0.74% (47.56%) |
17.11% (70.58%) |
| 総合 | 59.68% (76.62%) |
59.86% (76.56%) |
59.11% (74.94%) |
65.00% (101.04%) |
74.89% (69.41%) |
65.51% (82.32%) |
56.37% (74.69%) |
90.51% (73.54%) |
予想に反してSaraDでも検証出来た(笑) しかし、あと少しが伸びないのは変わらないなぁ。
0 件のコメント:
コメントを投稿