苦戦していたLightGBMのC APIでのテストがやっと出来る様になりました。いや、躓いていた原因は単に自分のポカで、ブースター名のコピペで別のブースター名にしてたのでリード違反が発生してました。
現時点で新馬戦CK用ブースターの最適RMSEは1.6836で
| 124R | 1点 | 芝(95R) | ダート(29R) | 8頭以下(19R) | 9～12頭(42R) | 13頭以上(63R) | 多点 |
| 単勝 | 26.61% (83.23%) |
27.37% (87.16%) |
24.14% (70.34%) |
26.32% (103.68%) |
21.43% (57.86%) |
30.16% (93.97%) |
58.87% (88.31%) |
| 複勝 | 50.81% (72.10%) |
53.68% (74.42%) |
41.38% (64.48%) |
57.89% (81.05%) |
42.86% (58.81%) |
53.97% (78.25%) |
86.29% (75.27%) |
| 枠連 | 10.75% (111.61%) |
15.15% (157.27%) |
0.00% (0.00%) |
-- (--) |
16.67% (162.14%) |
4.76% (56.67%) |
10.75% (118.50%) |
| 馬連 | 8.06% (86.94%) |
10.53% (113.47%) |
0.00% (0.00%) |
15.79% (97.89%) |
11.90% (125.24%) |
3.17% (58.10%) |
25.81% (126.85%) |
| ワイド | 18.55% (65.97%) |
21.05% (78.63%) |
10.34% (24.48%) |
36.84% (96.32%) |
21.43% (73.81%) |
11.11% (51.59%) |
43.55% (66.45%) |
| 馬単 | 4.03% (60.73%) |
5.26% (79.26%) |
0.00% (0.00%) |
5.26% (48.95%) |
7.14% (109.05%) |
1.59% (32.06%) |
25.81% (113.60%) |
| 三連複 | 2.42% (15.24%) |
3.16% (19.89%) |
0.00% (0.00%) |
15.79% (99.47%) |
0.00% (0.00%) |
0.00% (0.00%) |
11.29% (54.15%) |
| 三連単 | 0.00% (0.00%) |
0.00% (0.00%) |
0.00% (0.00%) |
0.00% (0.00%) |
0.00% (0.00%) |
0.00% (0.00%) |
11.29% (41.89%) |
| 総合 | 51.61% (60.37%) |
54.74% (73.05%) |
41.38% (20.09%) |
63.16% (75.34%) |
42.86% (73.36%) |
53.97% (46.33%) |
86.29% (67.01%) |
これに対してAutoMLでの新馬戦CK用学習モデルの最適RMSEは1.9544で
| 124R | 1点 | 芝(95R) | ダート(29R) | 8頭以下(19R) | 9～12頭(42R) | 13頭以上(63R) | 多点 |
| 単勝 | 25.00% (90.08%) |
26.32% (102.95%) |
20.69% (47.93%) |
31.58% (231.58%) |
21.43% (51.90%) |
25.40% (72.86%) |
45.97% (58.92%) |
| 複勝 | 53.23% (89.27%) |
54.74% (95.26%) |
48.28% (69.66%) |
57.89% (111.05%) |
47.62% (68.57%) |
55.56% (96.51%) |
83.87% (69.68%) |
| 枠連 | 12.90% (126.88%) |
16.67% (160.15%) |
3.70% (45.56%) |
-- (--) |
16.67% (126.19%) |
7.94% (103.17%) |
12.90% (74.20%) |
| 馬連 | 8.87% (84.84%) |
10.53% (80.42%) |
3.45% (99.31%) |
10.53% (40.00%) |
14.29% (129.52%) |
4.76% (68.57%) |
17.74% (56.80%) |
| ワイド | 16.94% (54.27%) |
17.89% (51.16%) |
13.79% (64.48%) |
42.11% (81.58%) |
21.43% (78.10%) |
6.35% (30.16%) |
33.87% (59.57%) |
| 馬単 | 4.03% (91.21%) |
4.21% (68.84%) |
3.45% (164.48%) |
5.26% (35.26%) |
4.76% (117.38%) |
3.17% (90.63%) |
17.74% (51.73%) |
| 三連複 | 4.84% (75.24%) |
4.21% (45.16%) |
6.90% (173.79%) |
10.53% (65.79%) |
4.76% (72.38%) |
3.17% (80.00%) |
12.10% (56.53%) |
| 三連単 | 1.61% (304.03%) |
1.05% (140.32%) |
3.45% (840.34%) |
0.00% (0.00%) |
2.38% (317.38%) |
1.59% (386.83%) |
12.10% (50.74%) |
| 総合 | 54.84% (114.08%) |
56.84% (90.37%) |
48.28% (189.43%) |
68.42% (80.75%) |
47.62% (120.18%) |
55.56% (116.09%) |
84.68% (54.97%) |
最適RMSEだけでは判断しづらいパターンだと思いますが、LightGBMはブースターをもう少しテストしていくと結果は違ってくる可能性がありそうです。通常用と順位予測のブースターも可能にしてトータルでの結果も出せる様にしたいですね。
追記 2024.10.18 7:09
夜勤明け帰宅後に昨日途中だった通常用検証メソッドを仕上げました。通常用CKブースターはまだ1つだけ試しに行ったもので最適RMSEは1.4225で
| 835R | 1点 | 芝(411R) | ダート(389R) | 障害(35R) | 8頭以下(62R) | 9～12頭(255R) | 13頭以上(518R) | 多点 |
| 単勝 | 25.87% (82.59%) |
26.28% (86.52%) |
24.94% (79.33%) |
31.43% (72.57%) |
37.10% (70.65%) |
31.37% (82.04%) |
21.81% (84.29%) |
51.62% (69.25%) |
| 複勝 | 57.60% (90.90%) |
55.96% (89.37%) |
58.10% (91.80%) |
71.43% (98.86%) |
66.13% (86.29%) |
66.27% (98.55%) |
52.32% (87.68%) |
86.47% (81.44%) |
| 枠連 | 13.51% (81.14%) |
12.80% (82.41%) |
14.05% (78.68%) |
15.79% (105.79%) |
-- (--) |
15.29% (75.22%) |
11.20% (75.44%) |
28.69% (77.12%) |
| 馬連 | 9.58% (62.86%) |
9.73% (64.31%) |
8.74% (61.13%) |
17.14% (65.14%) |
19.35% (53.06%) |
12.94% (72.43%) |
6.76% (59.32%) |
22.51% (69.85%) |
| ワイド | 23.95% (86.86%) |
22.87% (73.58%) |
23.65% (100.67%) |
40.00% (89.43%) |
46.77% (82.10%) |
29.80% (100.78%) |
18.34% (80.58%) |
43.95% (79.50%) |
| 馬単 | 5.27% (64.02%) |
4.38% (58.83%) |
5.40% (65.35%) |
14.29% (110.29%) |
11.29% (56.61%) |
7.45% (65.18%) |
3.47% (64.34%) |
22.51% (65.38%) |
| 三連複 | 5.39% (84.25%) |
6.08% (135.06%) |
4.11% (31.26%) |
11.43% (76.57%) |
17.74% (59.19%) |
8.24% (50.51%) |
2.51% (103.86%) |
14.73% (91.74%) |
| 三連単 | 1.32% (188.65%) |
1.46% (352.82%) |
0.77% (23.06%) |
5.71% (101.14%) |
4.84% (36.13%) |
2.35% (38.31%) |
0.39% (280.91%) |
14.73% (77.21%) |
| 総合 | 59.28% (92.86%) |
57.91% (118.69%) |
59.38% (66.31%) |
74.29% (89.02%) |
72.58% (63.43%) |
66.67% (72.88%) |
54.05% (104.55%) |
86.83% (76.39%) |
AutoMLでの最適RMSEは1.3376で
| 835R | 1点 | 芝(411R) | ダート(389R) | 障害(35R) | 8頭以下(62R) | 9～12頭(255R) | 13頭以上(518R) | 多点 |
| 単勝 | 24.79% (75.62%) |
23.84% (79.37%) |
24.42% (69.07%) |
40.00% (104.29%) |
41.94% (80.32%) |
29.41% (78.98%) |
20.46% (73.40%) |
53.29% (77.49%) |
| 複勝 | 55.81% (83.83%) |
51.09% (77.03%) |
59.90% (90.75%) |
65.71% (86.86%) |
61.29% (71.61%) |
60.78% (86.20%) |
52.70% (84.13%) |
89.58% (80.71%) |
| 枠連 | 13.37% (99.69%) |
12.50% (97.56%) |
13.77% (95.92%) |
21.05% (209.47%) |
-- (--) |
14.12% (92.39%) |
11.58% (92.70%) |
26.60% (79.52%) |
| 馬連 | 9.70% (68.13%) |
8.27% (56.93%) |
10.28% (78.48%) |
20.00% (84.57%) |
24.19% (77.10%) |
12.16% (70.55%) |
6.76% (65.87%) |
20.72% (64.64%) |
| ワイド | 21.08% (71.86%) |
21.41% (68.15%) |
19.28% (75.01%) |
37.14% (80.29%) |
50.00% (90.97%) |
23.14% (68.55%) |
16.60% (71.20%) |
41.80% (75.88%) |
| 馬単 | 4.79% (54.63%) |
4.14% (43.45%) |
4.37% (59.07%) |
17.14% (136.57%) |
14.52% (73.23%) |
6.27% (66.39%) |
2.90% (46.62%) |
20.72% (63.59%) |
| 三連複 | 5.63% (70.68%) |
6.33% (80.68%) |
4.37% (61.34%) |
11.43% (57.14%) |
14.52% (38.87%) |
7.06% (57.61%) |
3.86% (80.93%) |
14.97% (69.54%) |
| 三連単 | 1.56% (34.90%) |
1.70% (33.87%) |
1.29% (33.32%) |
2.86% (64.57%) |
3.23% (22.90%) |
2.35% (30.55%) |
0.97% (38.47%) |
14.97% (74.96%) |
| 総合 | 57.72% (69.39%) |
53.77% (66.42%) |
60.93% (70.16%) |
68.57% (96.52%) |
70.97% (65.00%) |
61.96% (68.90%) |
54.05% (69.17%) |
90.18% (73.26%) |
これ見ると随分と検討している感じはします。どちらも更に学習何度か試してみる価値は十分にありそうだし、順位予測も頑張って進めなきゃなんですが、これはそもそもこのブースターの結果をCSV出力して学習させなきゃなのでちょっと時間掛かりそう😔
0 件のコメント:
コメントを投稿