ノイズを削るって話と、精度を上げるって話。そんな中でちょっとJRA-VANの闇というか、まあ、時代の仕業なんだけど、そんなのを目の当たりにしてちょっと余分な時間を掛けてしまった。もうあまり気が進まない特徴量の追加なんですが、そろそろ今回を一旦区切りとしてってか、そもそもGWが本日までなので、朝から晩までコーディングな生活は終わりです。
まあ、そんなギリギリのGW最終日に新たな特徴量追加してCSV出力、まずはAutoMLで6時間学習して得た\(R^{2}\)値は0.2543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 | 26.11% (78.16%) |
24.99% (76.18%) |
26.97% (79.00%) |
30.00% (94.58%) |
34.70% (79.73%) |
28.06% (77.38%) |
24.68% (78.28%) |
55.66% (76.40%) |
| 複勝 | 58.12% (83.90%) |
57.67% (83.90%) |
57.90% (82.89%) |
67.50% (97.75%) |
70.78% (90.82%) |
63.29% (84.97%) |
55.25% (82.92%) |
88.57% (81.08%) |
| 枠連 | 14.05% (74.37%) |
12.85% (73.46%) |
14.42% (72.90%) |
25.25% (110.81%) |
-- (--) |
16.44% (76.71%) |
12.36% (68.88%) |
30.14% (77.34%) |
| 馬連 | 11.00% (68.10%) |
10.18% (62.04%) |
11.42% (73.83%) |
16.67% (74.92%) |
22.83% (72.51%) |
13.23% (69.79%) |
9.19% (67.14%) |
23.88% (68.25%) |
| ワイド | 25.33% (78.87%) |
25.10% (75.91%) |
25.15% (81.42%) |
30.83% (85.67%) |
45.21% (83.20%) |
30.04% (77.53%) |
21.96% (78.93%) |
46.80% (76.69%) |
| 馬単 | 6.11% (68.47%) |
5.39% (59.09%) |
6.50% (76.47%) |
10.83% (90.67%) |
10.96% (72.37%) |
7.05% (69.59%) |
5.36% (67.74%) |
23.88% (65.52%) |
| 三連複 | 6.22% (72.57%) |
7.10% (76.59%) |
5.47% (70.10%) |
4.17% (49.92%) |
16.44% (71.42%) |
8.28% (75.13%) |
4.61% (71.83%) |
16.58% (68.21%) |
| 三連単 | 1.74% (95.18%) |
1.95% (116.29%) |
1.58% (78.37%) |
0.83% (28.75%) |
4.11% (160.68%) |
2.60% (120.27%) |
1.24% (80.91%) |
16.58% (65.00%) |
| 総合 | 59.28% (77.50%) |
58.97% (78.03%) |
58.87% (76.89%) |
69.17% (78.42%) |
74.89% (90.10%) |
64.40% (81.42%) |
56.16% (74.58%) |
88.86% (68.59%) |
昨年を検証してみるとかなり頑張ってはいるんですが、もう一つ上に行ききれない。今はModel Builderで24時間学習させてるので、明日仕事終わった後の結果待ちです。
0 件のコメント:
コメントを投稿