諦めるとかはしたくないので

昨夜も試してたが書く気にもならない結果だった(笑) で、今晩は葉を増やして

NumberOfLeaves: 319
NumberOfIterations: 2000
LearningRate: 0.01
MaximumBinCountPerFeature: 31
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.8
FeatureFraction: 0.75
CustomGains: 0～500 = 0点, 501～1000は0.95乗に0～1000点

軽く10回試した結果、最適は

[Loop 3] NDCG@1 = 0.7671
[Loop 3] NDCG@3 = 0.7738
[Loop 3] NDCG@5 = 0.7889
[Loop 3] NDCG@10 = 0.8495

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.67% (72.77%)	24.41% (75.09%)	25.02% (71.46%)	23.60% (57.53%)	35.03% (72.80%)	26.92% (69.59%)	22.93% (73.94%)	54.79% (75.42%)
複勝	55.36% (79.00%)	53.36% (78.34%)	57.20% (79.88%)	58.43% (76.18%)	63.69% (86.31%)	60.30% (78.69%)	52.80% (78.47%)	89.58% (81.74%)
枠連	14.01% (77.43%)	13.42% (88.58%)	14.23% (64.67%)	18.99% (114.68%)	-- (--)	13.38% (58.17%)	13.42% (80.05%)	30.24% (77.99%)
馬連	10.27% (67.60%)	9.94% (78.18%)	10.46% (54.03%)	12.36% (105.06%)	22.93% (100.70%)	10.83% (60.20%)	8.95% (67.44%)	23.68% (76.05%)
ワイド	24.67% (74.45%)	23.71% (74.04%)	25.91% (74.93%)	21.35% (73.71%)	42.68% (80.32%)	27.37% (73.86%)	22.09% (74.16%)	46.82% (79.17%)
馬単	5.67% (60.30%)	5.09% (61.31%)	6.19% (55.17%)	6.74% (117.42%)	12.10% (71.34%)	6.62% (63.94%)	4.75% (57.98%)	23.68% (73.37%)
三連複	7.05% (108.99%)	6.89% (137.24%)	7.40% (85.78%)	4.49% (27.64%)	13.38% (103.82%)	7.67% (64.47%)	6.26% (126.01%)	16.90% (77.87%)
三連単	1.57% (109.22%)	1.17% (94.79%)	2.01% (129.69%)	1.12% (30.67%)	2.55% (161.53%)	2.11% (56.18%)	1.29% (124.36%)	16.90% (76.09%)
総合	57.32% (81.27%)	55.87% (85.90%)	58.65% (77.03%)	59.55% (74.80%)	66.88% (96.69%)	61.05% (65.64%)	55.09% (85.30%)	89.73% (76.49%)

ちょっと精度が上がった^^ 今週末はこの線で進めてみます。

追記 2026.10.23 19:49

NumberOfLeaves: 335
NumberOfIterations: 2000
LearningRate: 0.01
MaximumBinCountPerFeature: 31
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.8
FeatureFraction: 0.75
CustomGains: 0～500 = 0点, 501～1000は0.95乗に0～1000点

10回ためすと最適は

[Loop 5] NDCG@1 = 0.7672
[Loop 5] NDCG@3 = 0.7726
[Loop 5] NDCG@5 = 0.7891
[Loop 5] NDCG@10 = 0.8496

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.18% (70.65%)	23.94% (71.84%)	24.62% (71.00%)	21.35% (48.65%)	38.22% (84.39%)	26.62% (68.05%)	22.04% (70.41%)	55.25% (75.39%)
複勝	54.25% (76.81%)	51.88% (75.20%)	56.40% (78.54%)	58.43% (75.96%)	65.61% (88.85%)	58.35% (75.01%)	51.73% (76.43%)	89.50% (81.11%)
枠連	14.40% (79.35%)	13.33% (88.24%)	15.24% (70.78%)	16.46% (84.94%)	-- (--)	13.98% (65.19%)	13.70% (79.96%)	30.24% (77.07%)
馬連	10.50% (68.05%)	9.94% (77.82%)	11.02% (58.21%)	11.24% (65.39%)	21.66% (92.87%)	11.58% (63.76%)	9.12% (67.47%)	24.02% (79.97%)
ワイド	24.52% (73.65%)	23.94% (75.38%)	25.26% (72.77%)	22.47% (61.12%)	40.76% (78.28%)	27.22% (70.80%)	22.09% (74.30%)	46.70% (77.28%)
馬単	6.05% (59.64%)	5.79% (63.75%)	6.36% (56.19%)	5.62% (48.88%)	15.29% (96.94%)	7.52% (61.02%)	4.70% (55.86%)	24.02% (77.59%)
三連複	6.90% (108.30%)	6.89% (136.14%)	7.00% (80.29%)	5.62% (99.78%)	12.74% (85.99%)	7.67% (72.81%)	6.10% (123.46%)	16.59% (78.13%)
三連単	1.53% (93.02%)	1.33% (61.96%)	1.77% (129.42%)	1.12% (30.67%)	3.18% (131.40%)	2.11% (51.80%)	1.17% (104.98%)	16.59% (75.49%)
総合	56.36% (78.68%)	54.30% (81.16%)	58.25% (77.19%)	59.55% (64.13%)	68.79% (94.10%)	58.95% (66.05%)	54.31% (81.61%)	89.54% (76.78%)

これは精度ダウン。

Microsoft Rewards 33回目

ここに書くの忘れてました。昨日またamazonギフトに交換しました。前回から38日程度で大体ここの所のペースですね。40日ペースだと年9回程度って事で5,400円/年です。やり慣れてるので習慣になってて朝のちょっとした時間でやってます。他の買い物を経由とかしてって事ではないので悪くないと思い継続してます。

Microsoft Rewards 紹介リンク

更にランキング学習の続き

NumberOfLeaves: 303
NumberOfIterations: 3000
LearningRate: 0.01
MaximumBinCountPerFeature: 31
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.8
FeatureFraction: 0.75
CustomGains: 0～500 = 0点, 501～1000は0.95乗に0～1000点

昨夜これを40回スタートして寝た。最適は

[Loop 15] NDCG@1 = 0.7671
[Loop 15] NDCG@3 = 0.7730
[Loop 15] NDCG@5 = 0.7887
[Loop 15] NDCG@10 = 0.8498

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.37% (73.60%)	23.87% (74.30%)	25.10% (74.60%)	21.35% (49.66%)	33.76% (72.36%)	25.41% (62.20%)	23.15% (77.95%)	54.87% (75.76%)
複勝	55.40% (79.59%)	53.21% (78.12%)	57.36% (81.17%)	59.55% (78.76%)	64.33% (86.43%)	59.55% (76.86%)	53.08% (80.01%)	89.62% (82.11%)
枠連	14.27% (81.66%)	13.05% (88.24%)	15.33% (74.75%)	15.19% (94.43%)	-- (--)	13.23% (61.67%)	13.81% (84.30%)	30.28% (76.86%)
馬連	10.42% (71.71%)	9.70% (76.28%)	11.18% (66.68%)	10.11% (76.40%)	21.02% (88.41%)	11.13% (68.23%)	9.23% (71.54%)	23.72% (79.10%)
ワイド	24.75% (77.06%)	23.94% (76.75%)	25.82% (78.33%)	21.35% (63.82%)	38.85% (76.88%)	27.67% (76.08%)	22.43% (77.44%)	47.24% (79.88%)
馬単	5.86% (63.75%)	5.09% (63.04%)	6.76% (64.47%)	4.49% (63.82%)	10.83% (85.03%)	6.17% (50.68%)	5.31% (66.74%)	23.72% (76.64%)
三連複	6.90% (108.40%)	6.65% (131.21%)	7.24% (85.57%)	5.62% (99.78%)	15.92% (122.04%)	7.37% (66.87%)	5.93% (122.65%)	17.20% (78.74%)
三連単	1.69% (126.79%)	1.25% (96.03%)	2.25% (167.50%)	0.00% (0.00%)	4.46% (214.59%)	1.80% (48.29%)	1.40% (148.28%)	17.20% (75.14%)
総合	57.16% (85.37%)	55.40% (85.44%)	58.73% (86.71%)	60.67% (65.43%)	67.52% (106.53%)	59.70% (63.86%)	55.31% (91.12%)	89.81% (76.72%)

次点で

[Loop 2] NDCG@1 = 0.7666
[Loop 2] NDCG@3 = 0.7735
[Loop 2] NDCG@5 = 0.7890
[Loop 2] NDCG@10 = 0.8500

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.21% (71.43%)	23.08% (70.05%)	25.50% (74.26%)	22.47% (51.91%)	33.12% (71.53%)	26.47% (68.20%)	22.60% (72.63%)	55.21% (75.31%)
複勝	54.87% (77.79%)	52.11% (75.38%)	57.44% (80.50%)	58.43% (74.72%)	64.33% (84.20%)	58.95% (76.63%)	52.52% (77.66%)	89.54% (82.13%)
枠連	14.18% (82.58%)	13.05% (87.02%)	14.90% (77.11%)	18.99% (103.16%)	-- (--)	14.59% (66.51%)	13.20% (83.71%)	31.01% (78.12%)
馬連	10.15% (67.03%)	9.47% (72.61%)	10.70% (60.72%)	12.36% (75.06%)	22.93% (95.29%)	11.73% (65.16%)	8.45% (65.24%)	23.98% (77.51%)
ワイド	24.02% (72.62%)	23.24% (73.81%)	25.02% (72.44%)	21.35% (58.20%)	41.40% (80.38%)	26.92% (71.35%)	21.42% (72.42%)	47.47% (80.83%)
馬単	5.40% (53.33%)	4.46% (48.68%)	6.36% (58.43%)	5.62% (48.88%)	12.74% (97.13%)	6.92% (60.33%)	4.19% (46.88%)	23.98% (75.17%)
三連複	7.24% (107.98%)	6.96% (134.69%)	7.64% (85.82%)	5.62% (33.71%)	15.29% (102.99%)	8.27% (67.47%)	6.15% (123.48%)	17.20% (76.47%)
三連単	1.61% (90.43%)	1.10% (42.57%)	2.17% (143.92%)	1.12% (30.67%)	5.10% (190.25%)	1.80% (51.62%)	1.23% (96.10%)	17.20% (75.38%)
総合	57.01% (77.84%)	54.69% (75.38%)	59.21% (81.68%)	59.55% (58.92%)	68.15% (103.11%)	60.00% (65.91%)	54.92% (79.76%)	89.85% (76.47%)

次はどう攻めるかだな^^;

8月のランキング学習を真似て

８月終わりにここに書いてたランキングモデルはここ最近のより精度が良かった記憶なので試してみた。当時と違うのは、学習データ期間が2024年までだったのが2025年までに増えた。

NumberOfLeaves: 303
NumberOfIterations: 2000
LearningRate: 0.01
MaximumBinCountPerFeature: 31
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.8
FeatureFraction: 0.75
CustomGains: 0～500 = 0点, 501～1000は0.95乗に0～1000点

その時は40回と書かれてるが取敢えず10回やると

[Loop 10] NDCG@1 = 0.7672
[Loop 10] NDCG@3 = 0.7727
[Loop 10] NDCG@5 = 0.7883
[Loop 10] NDCG@10 = 0.8493

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.10% (70.65%)	23.16% (69.30%)	25.18% (73.40%)	22.47% (51.57%)	33.12% (68.85%)	25.86% (65.44%)	22.65% (72.75%)	55.33% (77.05%)
複勝	55.25% (78.93%)	52.66% (76.79%)	57.52% (81.01%)	60.67% (80.56%)	64.33% (88.22%)	59.55% (77.43%)	52.85% (78.67%)	89.62% (81.15%)
枠連	14.18% (80.48%)	12.87% (87.11%)	15.41% (75.39%)	13.92% (65.19%)	-- (--)	14.14% (63.01%)	13.37% (82.25%)	30.96% (78.89%)
馬連	10.19% (69.37%)	9.31% (76.52%)	11.18% (63.56%)	8.99% (47.98%)	21.02% (90.64%)	11.43% (61.92%)	8.78% (70.27%)	23.91% (73.31%)
ワイド	24.87% (76.23%)	23.94% (75.76%)	26.15% (77.35%)	20.22% (67.30%)	41.40% (77.58%)	28.42% (77.26%)	22.09% (75.72%)	46.82% (75.20%)
馬単	5.82% (60.96%)	4.69% (56.32%)	6.92% (65.99%)	6.74% (57.30%)	11.46% (63.63%)	6.92% (55.13%)	4.92% (62.89%)	23.91% (72.16%)
三連複	6.82% (73.74%)	6.34% (66.03%)	7.48% (84.97%)	4.49% (27.64%)	13.38% (63.82%)	7.37% (61.74%)	6.04% (79.07%)	16.55% (74.96%)
三連単	1.61% (102.80%)	1.02% (74.78%)	2.25% (136.78%)	1.12% (30.67%)	2.55% (30.13%)	1.80% (44.33%)	1.45% (130.93%)	16.55% (73.43%)
総合	57.13% (76.60%)	54.85% (72.56%)	59.13% (82.35%)	61.80% (53.36%)	66.88% (68.98%)	60.45% (63.28%)	55.03% (81.57%)	89.77% (74.49%)

やはり的中率はここ最近のモデルより良い。配収率は負けてますけど^^; まっ、試しに10回だし、イテレーション数も2,000と抑えた感じなので、ここ最近の3,000で40回スタートさせてみるかな。

一応今回のNDCG@3的には

[Loop 7] NDCG@1 = 0.7666
[Loop 7] NDCG@3 = 0.7730
[Loop 7] NDCG@5 = 0.7888
[Loop 7] NDCG@10 = 0.8495

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.52% (72.59%)	24.26% (73.83%)	24.78% (72.15%)	24.72% (60.90%)	31.85% (67.71%)	27.07% (67.61%)	22.93% (74.87%)	55.33% (76.32%)
複勝	55.63% (79.28%)	53.68% (78.33%)	57.36% (80.37%)	59.55% (77.87%)	64.33% (89.43%)	60.75% (79.20%)	52.96% (78.42%)	89.73% (82.01%)
枠連	13.93% (75.20%)	12.87% (86.37%)	14.73% (64.72%)	16.46% (78.10%)	-- (--)	12.93% (56.21%)	13.48% (77.85%)	30.66% (76.85%)
馬連	10.08% (63.41%)	9.39% (72.80%)	10.78% (54.50%)	10.11% (52.92%)	20.38% (90.57%)	10.68% (55.14%)	8.95% (64.09%)	24.21% (79.66%)
ワイド	24.06% (72.49%)	23.24% (73.30%)	24.94% (71.17%)	23.60% (79.44%)	38.85% (71.40%)	27.22% (73.29%)	21.59% (72.29%)	47.39% (78.65%)
馬単	5.67% (54.59%)	5.16% (59.07%)	6.11% (50.23%)	6.74% (51.24%)	12.10% (91.08%)	6.77% (54.35%)	4.70% (51.48%)	24.21% (77.17%)
三連複	6.93% (103.52%)	6.96% (135.84%)	7.00% (70.56%)	5.62% (99.78%)	14.65% (102.55%)	7.82% (71.52%)	5.93% (115.51%)	16.32% (69.37%)
三連単	1.38% (60.31%)	1.17% (60.16%)	1.61% (62.57%)	1.12% (30.67%)	2.55% (146.75%)	1.95% (45.53%)	1.06% (58.21%)	16.32% (66.71%)
総合	57.51% (72.64%)	56.18% (79.84%)	58.73% (65.79%)	59.55% (66.20%)	67.52% (94.21%)	61.20% (62.86%)	55.26% (74.09%)	89.81% (71.83%)

なんとも言えない(笑)

ランキング学習の続き

ランキング学習を継続してました。葉の数を増やして

NumberOfLeaves: 511
NumberOfIterations: 3000
LearningRate: 0.01
MaximumBinCountPerFeature: 127
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.8
FeatureFraction: 0.75
CustomGains: 0～300 = 0点, 301～1000は0.95乗に0～1000点

昨夜就寝前に40回スタート。現在38回目が動いてますが、ここまでのベストは

[Loop 14] NDCG@1 = 0.7688
[Loop 14] NDCG@3 = 0.7741
[Loop 14] NDCG@5 = 0.7903
[Loop 14] NDCG@10 = 0.8509

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.18% (75.42%)	23.63% (74.91%)	24.70% (76.42%)	24.72% (68.65%)	33.12% (71.02%)	26.32% (68.98%)	22.60% (78.20%)	54.94% (76.22%)
複勝	54.41% (78.52%)	52.58% (77.54%)	55.91% (79.45%)	59.55% (79.55%)	62.42% (82.93%)	58.95% (76.65%)	52.01% (78.83%)	89.20% (81.53%)
枠連	14.35% (82.20%)	13.05% (85.78%)	15.16% (75.50%)	20.25% (132.91%)	-- (--)	12.48% (51.40%)	14.21% (88.83%)	30.62% (76.34%)
馬連	10.23% (67.84%)	9.23% (66.31%)	11.10% (66.81%)	12.36% (104.16%)	19.11% (73.89%)	10.38% (55.26%)	9.40% (71.99%)	24.06% (75.94%)
ワイド	23.68% (73.77%)	22.30% (68.90%)	25.26% (78.24%)	21.35% (81.24%)	36.31% (65.10%)	26.02% (68.95%)	21.70% (76.33%)	46.90% (78.81%)
馬単	5.98% (71.51%)	5.09% (64.87%)	6.76% (75.00%)	7.87% (118.09%)	10.19% (61.91%)	6.32% (55.80%)	5.48% (78.19%)	24.06% (73.31%)
三連複	6.82% (77.29%)	6.49% (71.49%)	7.24% (84.39%)	5.62% (61.24%)	13.38% (89.49%)	6.62% (59.23%)	6.32% (82.93%)	16.74% (83.34%)
三連単	1.69% (114.31%)	1.17% (90.38%)	2.17% (141.21%)	2.25% (82.13%)	3.82% (181.27%)	1.50% (40.42%)	1.57% (135.91%)	16.74% (79.46%)
総合	56.51% (80.08%)	55.01% (74.82%)	57.76% (84.69%)	60.67% (90.40%)	65.61% (89.37%)	59.25% (59.59%)	54.70% (86.40%)	89.43% (78.53%)

これ、回収率が上がってるけど的中率的には微妙に悪い感じだ。厳密にはワイドが低いんだよね。ワイド的にはNDCG@3が意味あると思うけど、0.7741はベストでもある。まあ、後は@5や@10的にも見た方が良いのかもだけど。

新馬戦だけだと

183R	1点	芝(129R)	ダート(54R)	8頭以下(28R)	9～12頭(68R)	13頭以上(87R)	多点
単勝	20.22% (92.90%)	23.26% (82.09%)	12.96% (118.70%)	32.14% (68.57%)	26.47% (73.24%)	11.49% (116.09%)	46.99% (74.08%)
複勝	44.81% (75.41%)	47.29% (71.40%)	38.89% (85.00%)	53.57% (73.93%)	48.53% (66.62%)	39.08% (82.76%)	80.33% (71.69%)
枠連	10.88% (127.35%)	14.43% (176.49%)	4.00% (32.00%)	-- (--)	11.76% (69.26%)	9.20% (161.03%)	21.77% (78.95%)
馬連	8.74% (67.32%)	11.63% (87.91%)	1.85% (18.15%)	21.43% (157.86%)	8.82% (58.97%)	4.60% (44.71%)	18.58% (118.32%)
ワイド	15.30% (52.51%)	18.60% (55.74%)	7.41% (44.81%)	35.71% (100.00%)	16.18% (44.41%)	8.05% (43.56%)	33.88% (68.91%)
馬単	5.46% (63.01%)	6.98% (81.94%)	1.85% (17.78%)	14.29% (177.86%)	7.35% (72.50%)	1.15% (18.62%)	18.58% (106.71%)
三連複	2.19% (52.08%)	3.10% (73.88%)	0.00% (0.00%)	10.71% (293.21%)	1.47% (19.41%)	0.00% (0.00%)	6.56% (32.60%)
三連単	1.09% (121.97%)	1.55% (173.02%)	0.00% (0.00%)	7.14% (797.14%)	0.00% (0.00%)	0.00% (0.00%)	6.56% (20.47%)
総合	45.90% (80.41%)	48.84% (97.87%)	38.89% (39.63%)	57.14% (238.37%)	48.53% (50.55%)	40.23% (58.35%)	80.33% (50.53%)

それ以外だと

2,427R	1点	芝(1,149R)	ダート(1,189R)	障害(89R)	8頭以下(129R)	9～12頭(597R)	13頭以上(1,701R)	多点
単勝	24.47% (74.10%)	23.67% (74.11%)	25.23% (74.50%)	24.72% (68.65%)	33.33% (71.55%)	26.30% (68.49%)	23.16% (76.26%)	55.54% (76.38%)
複勝	55.13% (78.75%)	53.18% (78.22%)	56.69% (79.20%)	59.55% (79.55%)	64.34% (84.88%)	60.13% (77.79%)	52.67% (78.62%)	89.86% (82.27%)
枠連	14.58% (79.18%)	12.92% (76.90%)	15.65% (77.43%)	20.25% (132.91%)	-- (--)	12.56% (49.36%)	14.46% (85.13%)	31.21% (76.16%)
馬連	10.34% (67.88%)	8.96% (63.89%)	11.52% (69.02%)	12.36% (104.16%)	18.60% (55.66%)	10.55% (54.84%)	9.64% (73.39%)	24.47% (72.74%)
ワイド	24.31% (75.37%)	22.72% (70.38%)	26.07% (79.76%)	21.35% (81.24%)	36.43% (57.52%)	27.14% (71.74%)	22.40% (78.00%)	47.88% (79.56%)
馬単	6.02% (72.15%)	4.87% (62.95%)	6.98% (77.60%)	7.87% (118.09%)	9.30% (36.74%)	6.20% (53.90%)	5.70% (81.24%)	24.47% (70.79%)
三連複	7.17% (79.19%)	6.88% (71.23%)	7.57% (88.23%)	5.62% (61.24%)	13.95% (45.27%)	7.20% (63.77%)	6.64% (87.17%)	17.51% (87.16%)
三連単	1.73% (113.73%)	1.13% (81.10%)	2.27% (147.63%)	2.25% (82.13%)	3.10% (47.60%)	1.68% (45.03%)	1.65% (142.86%)	17.51% (83.91%)
総合	57.31% (80.05%)	55.70% (72.27%)	58.62% (86.73%)	60.67% (90.40%)	67.44% (57.03%)	60.47% (60.62%)	55.44% (87.83%)	90.11% (80.63%)

もう少し改善したいな^^;

第31回秋華賞

ちょっとカーオーディオの取付で時間が取れず、ざっくりと独自予想してみた。

土つかずの1番人気は強そうだけど、距離伸びて飛んでもらい2,000mの実績で狙ってみました。

さて、カーオーディオの作業に戻ります。

Pioneer MVH-5600

ポンコツマイカーのナビが逝きました。まあね、HDDナビなので20年も動く事が奇跡レベルなので諦めました。ポイント使ってPioneer MVH-5600を購入。DIYでって初めてやったんですが、ステーが無い😭 これ、近所のイエローハットとかに行くとあるんだろうか? 元々2DINのストラーダが付いてたので1DINでは下が空くのは分かってたのでAD-378って小物入れも一緒に購入して有ったんだが、ステーが無いとは想定してなかった。まあ、そこは素人だね(笑)

午前中の到着して作業したんだがってか、既に昨日ナビは外しておいた。ハーネス関連は流用出来たっぽいけど、多分、配線が完全じゃないのでなのか音が出てない。やり直さなきゃです。

Pioneer MVH-5600 12,164円
AD-378 1,082円

追記 2026.10.19 18:10

流石に素人が初めてDIYしてるので分からない事や諸々で昨日はイエローハットに出向いて使えそうなの有るか見回すがそれっぽいのが無い。30年ぶり位にオートバックスに行って駐車場に車止めて歩き出すと、駐車場の係の人が「お客さん! 入り口はこっちだよ!」って感じで全く把握出来てなかった(笑) で、店内見回してるとトヨタ サイドパネルとやらを見つけ878円で購入して取付を試すがやはりステーが無いのでうまく固定出来ない。

本日大腸カメラ検査後に帰宅し、ふと外したストラーダを見ると、使えそうなブラケット発見! 外して使ってみると、ちょっと若干無理やりな感じもあるが固定出来た。やれやれです。後何年この車に乗るかは不明だけど、それまでは無事に動いてくれれば良いや^^

ランキング学習のメモ

先日のリリースに添付したランキング学習のモデルはRankCK3000_loop1_NDCG7679_7741.zipで

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.75% (75.89%)	25.04% (81.96%)	24.70% (71.30%)	21.35% (52.81%)	33.76% (74.78%)	27.07% (71.82%)	23.10% (77.51%)	54.67% (75.35%)
複勝	55.52% (80.09%)	53.36% (78.44%)	57.52% (82.08%)	58.43% (76.07%)	64.33% (92.36%)	60.45% (78.06%)	52.91% (79.77%)	89.66% (81.16%)
枠連	13.76% (74.67%)	12.13% (75.46%)	14.90% (72.29%)	18.99% (99.24%)	-- (--)	13.98% (60.81%)	12.86% (75.44%)	30.49% (76.91%)
馬連	9.96% (65.54%)	8.69% (66.33%)	11.10% (64.22%)	12.36% (72.70%)	20.38% (87.83%)	11.43% (63.91%)	8.50% (64.19%)	23.60% (71.87%)
ワイド	24.18% (73.08%)	22.69% (71.83%)	25.91% (75.45%)	21.35% (58.09%)	40.13% (78.85%)	27.22% (70.56%)	21.64% (73.52%)	46.17% (76.18%)
馬単	5.48% (56.70%)	4.93% (58.16%)	6.11% (56.11%)	4.49% (43.93%)	12.10% (81.21%)	6.77% (66.80%)	4.42% (50.79%)	23.60% (70.33%)
三連複	6.59% (71.77%)	6.34% (68.12%)	7.00% (78.69%)	4.49% (27.64%)	13.38% (90.45%)	7.37% (61.35%)	5.70% (74.01%)	16.93% (77.84%)
三連単	1.46% (90.74%)	1.02% (49.79%)	1.93% (137.14%)	1.12% (30.67%)	1.91% (106.43%)	1.65% (58.74%)	1.34% (101.26%)	16.93% (74.78%)
総合	57.20% (73.55%)	55.56% (68.63%)	58.81% (79.71%)	58.43% (57.05%)	67.52% (87.42%)	60.90% (66.51%)	54.92% (74.56%)	89.96% (74.93%)

この際に使ったパラメーターが分からなくなった(笑) イテレーション数3,000なのは分かるが、率は多分0.01だったのかな^^; モデルサイズが21MBなのでそんな気がするけど。

パラメータ / 指標	条件 1	条件 2	条件 3	条件 4	条件 5	条件 6	条件 7
NumberOfLeaves	79	95	95	111	127	255	351
NumberOfIterations	3,000	3,000	3,000	3,000	3,000	3,000	3,000
LearningRate	0.010	0.010	0.005	0.010	0.010	0.010	0.010
MaximumBin	127	127	127	127	127	127	127
MinimumExample	50	50	50	50	50	50	50
SubsampleFraction	0.800	0.800	0.800	0.800	0.800	0.800	0.800
FeatureFraction	0.75	0.75	0.75	0.75	0.75	0.75	0.75
ボーダー	300	300	300	300	300	300	300
乗数	0.95	0.95	0.95	0.95	0.95	0.95	0.95
NDCG@1	0.7667	0.7660	0.7635	0.7673	0.7669	0.7668	0.7665
NDCG@3	0.7725	0.7723	0.7701	0.7730	0.7731	0.7731	0.7736
NDCG@5	0.7889	0.7891	0.7867	0.7893	0.7895	0.7893	0.7895
NDCG@10	0.8498	0.8499	0.8482	0.8502	0.8504	0.8505	0.8507

なんか以前、葉を増やして出したモデルっぽいな。最近の走破タイムでは葉の数を絞ってたので忘れてるけど300以上も試してた記憶もあるからなぁ。

NumberOfLeaves: 335
NumberOfIterations: 3000
LearningRate: 0.01
MaximumBinCountPerFeature: 127
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.8
FeatureFraction: 0.75
CustomGains: 0～300 = 0点, 301～1000は0.95乗に0～1000点

これ10回試したら最後に

[Loop 10] NDCG@1 = 0.7683
[Loop 10] NDCG@3 = 0.7740
[Loop 10] NDCG@5 = 0.7902
[Loop 10] NDCG@10 = 0.8509

が出た。

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.37% (74.99%)	23.87% (74.20%)	25.02% (77.40%)	22.47% (52.58%)	34.39% (74.39%)	27.82% (71.64%)	22.20% (76.29%)	54.90% (75.18%)
複勝	54.90% (78.89%)	52.66% (76.56%)	57.12% (81.78%)	56.18% (72.02%)	64.33% (85.92%)	59.70% (76.42%)	52.29% (79.19%)	89.77% (81.65%)
枠連	13.67% (76.25%)	12.13% (81.24%)	14.73% (69.86%)	18.99% (103.16%)	-- (--)	13.38% (58.06%)	12.98% (78.54%)	30.28% (78.02%)
馬連	10.04% (64.96%)	9.00% (64.98%)	10.94% (64.22%)	12.36% (75.06%)	21.02% (80.89%)	10.83% (61.10%)	8.78% (64.99%)	23.52% (74.94%)
ワイド	24.44% (73.08%)	23.47% (72.16%)	25.66% (75.07%)	21.35% (58.43%)	40.76% (77.45%)	27.37% (72.23%)	21.92% (73.01%)	46.51% (79.31%)
馬単	5.56% (59.32%)	4.69% (49.66%)	6.36% (69.24%)	6.74% (59.44%)	12.10% (68.79%)	6.92% (62.26%)	4.47% (57.40%)	23.52% (72.10%)
三連複	7.05% (77.72%)	7.12% (78.04%)	7.16% (80.98%)	4.49% (27.64%)	15.29% (116.94%)	7.52% (64.41%)	6.15% (79.23%)	16.36% (76.34%)
三連単	1.61% (93.76%)	1.25% (57.75%)	2.01% (135.30%)	1.12% (30.67%)	3.82% (181.27%)	1.95% (49.79%)	1.29% (102.43%)	16.36% (71.54%)
総合	56.67% (74.85%)	54.85% (69.10%)	58.49% (81.81%)	57.30% (59.26%)	68.15% (97.95%)	60.30% (64.49%)	54.31% (76.38%)	90.04% (73.89%)

やはり、この辺りなんだなぁ^^;

SQLite Version 3.54.1

1週間で新たなリリースがありました。前回の不具合報告から"Fixed"との事なので入れてみました。

フルセットアップは前回も問題なかったですが、前回よりちょっと早い? その後、最適化もしたんですが、何となくこれも早くなった気がする。で、問題だった検証も、ヤバい位早くなったと思う。これ、キャッシュとかの影響もきになるけど、単にフルセットアップ→最適化→検証と行っただけなので、キャッシュはされている気がしないのに^^; ちょっと信じ難いレベルの改善がされてる!

しかし、そもそも早いSQLiteなんだがまだまだ改善されるとか驚かされます。

メモるの忘れてた

サボってた訳ではなく単純にここに書くのを忘れてた^^; しかもどんなパラメーター試してたのかも忘れた(笑) 学習させたモデルは残ってるのでRMSEとかはファイル名で分かるけど...しかし、SQLiteをバージョンアップしたのが原因なのか、まあ、そもそも最初の検証時はデータがキャッシュされてないってもあって遅いんだが、それにしても異常に遅いorz 確か数分で終わるレベルの筈なんだが、そもそも1か月がまだ終わらないのは酷いな。

まさかとは思ったけど、SQLite Version 3.54.0にしたら激遅になったっぽい😱 今年の1月～9月の検証で１時間以上たっても3月にやっと入った程度だったのが、元に戻したら６分２８秒で検証完了(笑) で、ああっ、通常用走破タイム学習RMSE 1.5893は

2,427R	1点	芝(1,149R)	ダート(1,189R)	障害(89R)	8頭以下(129R)	9～12頭(597R)	13頭以上(1,701R)	多点
単勝	20.19% (69.93%)	20.19% (75.27%)	20.86% (66.78%)	11.24% (42.92%)	28.68% (77.29%)	20.44% (57.77%)	19.46% (73.63%)	49.57% (76.03%)
複勝	50.60% (81.03%)	47.00% (78.19%)	53.83% (83.41%)	53.93% (85.84%)	58.91% (85.66%)	51.93% (77.44%)	49.50% (81.94%)	85.41% (80.63%)
枠連	12.13% (78.91%)	11.50% (85.28%)	12.73% (73.76%)	11.39% (72.78%)	-- (--)	11.56% (62.73%)	11.64% (80.09%)	27.71% (83.80%)
馬連	8.45% (70.66%)	7.83% (79.63%)	9.00% (61.93%)	8.99% (71.46%)	15.50% (72.64%)	8.21% (58.31%)	8.00% (74.84%)	20.27% (77.86%)
ワイド	21.14% (77.61%)	18.62% (71.32%)	23.63% (81.48%)	20.22% (107.08%)	33.33% (62.25%)	23.45% (82.11%)	19.40% (77.20%)	41.20% (79.85%)
馬単	4.08% (63.07%)	3.39% (65.38%)	5.05% (65.57%)	0.00% (0.00%)	7.75% (76.98%)	4.19% (46.68%)	3.76% (67.77%)	20.27% (76.07%)
三連複	5.27% (97.31%)	4.26% (107.21%)	6.31% (92.54%)	4.49% (33.26%)	9.30% (37.83%)	6.03% (80.74%)	4.70% (107.64%)	14.09% (71.25%)
三連単	0.70% (61.07%)	0.52% (101.88%)	0.93% (26.21%)	0.00% (0.00%)	1.55% (56.28%)	1.01% (42.45%)	0.53% (67.97%)	14.09% (67.48%)
総合	52.25% (74.90%)	49.26% (82.98%)	55.00% (68.93%)	53.93% (51.37%)	63.57% (66.99%)	53.43% (63.53%)	50.97% (78.89%)	85.74% (72.46%)

まだまだだけど少しは改善した。しかし、何してこうなったのか全く忘れてるので次の手が分からないorz

追記 2026.10.10 22:19

NumberOfLeaves: 95
NumberOfIterations: 1,500
LearningRate: 0.020
MaximumBinCountPerFeature: 127
MinimumExampleCountPerLeaf: 20
SubsampleFraction: 0.075
FeatureFraction: 0.980
L2Regularization: 0.500

これでRMSE 1.5878

2,427R	1点	芝(1,149R)	ダート(1,189R)	障害(89R)	8頭以下(129R)	9～12頭(597R)	13頭以上(1,701R)	多点
単勝	20.19% (72.18%)	19.67% (75.90%)	21.11% (69.55%)	14.61% (59.44%)	25.58% (64.11%)	21.78% (66.77%)	19.22% (74.70%)	49.77% (75.37%)
複勝	51.17% (81.62%)	48.13% (81.06%)	54.25% (82.84%)	49.44% (72.47%)	58.91% (88.60%)	55.95% (83.05%)	48.91% (80.59%)	86.65% (81.91%)
枠連	12.27% (77.44%)	11.10% (86.29%)	13.62% (72.12%)	7.59% (42.66%)	-- (--)	11.06% (64.17%)	11.99% (77.68%)	27.81% (86.55%)
馬連	8.74% (69.35%)	7.92% (76.23%)	9.84% (65.00%)	4.49% (38.65%)	16.28% (79.30%)	7.87% (58.74%)	8.47% (72.32%)	20.56% (79.45%)
ワイド	21.63% (78.38%)	18.80% (73.37%)	24.31% (80.95%)	22.47% (108.88%)	30.23% (54.96%)	23.95% (83.79%)	20.16% (78.27%)	41.94% (82.40%)
馬単	4.37% (64.96%)	3.92% (72.42%)	5.13% (62.61%)	0.00% (0.00%)	7.75% (57.60%)	4.19% (46.72%)	4.17% (71.92%)	20.56% (76.95%)
三連複	5.81% (88.59%)	5.13% (103.20%)	6.56% (78.60%)	4.49% (33.26%)	12.40% (57.29%)	6.87% (88.49%)	4.94% (90.99%)	14.50% (88.29%)
三連単	0.74% (45.01%)	0.70% (57.29%)	0.84% (36.52%)	0.00% (0.00%)	2.33% (53.72%)	0.67% (40.94%)	0.65% (45.78%)	14.50% (85.54%)
総合	52.41% (72.13%)	49.52% (78.08%)	55.34% (68.50%)	50.56% (44.44%)	60.47% (65.08%)	57.45% (66.58%)	50.03% (74.03%)	86.98% (83.34%)

この辺りがそろそろ限界に近いのかも。いや、現行ML.NETではね。来月に期待してます。

SQLite Version 3.54.0

ここしばらく動きが無かったけど、昨日(10/9)バージョンアップされたので早速入れてみた。本来はSourceGear.sqlite3を待つのですが、この更新タイミングが毎回微妙なので先行してSQLite本体のprecompiledダウンロードして適用。

で、早速フルセットアップさせてみた。

公式アナウンスでは4%改善との事。こちらでは、フルセットアップ時に毎回微妙にバックグラウンドで他の処理させてたり、そもそも時期が違えば期間も長くなるので正確な比較は出来ないけど、まあ、33分は速いのかな(笑)

追記 2026.10.10 11:41

セットアップって基本データの書込みばかりやってるだけなんだが、検証走らせると激遅でした。Version 3.53.4に戻しました。

追記 2026.10.14 20:06

SQLiteフォーラムに報告して既に修正済みとの事だけど、独自ビルドすれば修正されるんだろうけど、一旦様子見します。

ちょっとした言い訳(笑)

別に手を抜いてるとか、気力が失せた訳じゃないんです。やれるだけやるって気持ちは未だあるんです。しかし、ここまで来ると、11月10日にリリース予定の.NET11がどうしても気になる。ってのも、そもそもその段階でML.NET 6.0も正式リリース(現時点ではpreview)なり、もしかするとLightGBMも4.7が採用される。更に、元々.NET10で既に何度かリリースしてたのに、Visual Studioの.NET10の発行時のバグで.NET9にダウングレードを余儀なくされた。

モデルの学習は時間が掛かるのは周知の事実。一度はLightGBM 4.7でやってたのを現時点では4.6にもどしてる。この状態でモデルの学習を突き詰めても新たにML.NET 6.0の正式リリース及びLightGBM 4.7がって事になれば、全て最初からやり直しになる。

まあ、これら以外にも微妙に気になる.NET11からの新機能にも期待はしてます。なので、年内に大幅なアップグレードした物がリリース出来る様に今はちょっとだけ、その下準備(笑)

いやね、AIに何度も騙されたから、ローカルLLMとかって事でここ数日試しました。OllamaとかLM Studioとかね。Ollamaは8年前に組んだこのPCでは動きませんorz いや、CUDA関連かもなんだが、NVIDIAのドライバーの更新で何度か沼に嵌ってOSのクリーンインストールしてるので、試す勇気がない。更にLM Studio Bionicはグラボが古いのでもっさりした反応だし、そもそも学習されてるデータが若干古い。

サボるつもりは全く無いです。限られた時間でやれる事はやります😉

葉の数抑えた通常用走破タイム学習

葉の数を抑えると何度か回してってのは無意味っぽい。だいたい1回目が一番良い(笑)

パラメータ / 指標	条件 1	条件 2	条件 3	条件 4	条件 5	条件 6	条件 7
NumberOfLeaves	63	79	47	63	63	63	63
NumberOfIterations	1,500	1,500	1,500	1,500	1,500	1,500	1,500
LearningRate	0.020	0.020	0.020	0.020	0.020	0.020	0.020
MaximumBin	63	63	63	79	95	111	127
MinimumExample	50	50	50	50	50	50	50
SubsampleFraction	0.050	0.050	0.050	0.050	0.050	0.050	0.050
FeatureFraction	0.800	0.800	0.800	0.800	0.800	0.800	0.800
L2Regularization	0.500	0.500	0.500	0.500	0.500	0.500	0.500
RMSE	1.6048	1.6072	1.6217	1.6167	1.6073	1.6205	1.6061

追記 2026.10.4 21:22

NumberOfLeaves: 79
NumberOfIterations: 1,500
LearningRate: 0.020
MaximumBinCountPerFeature: 127
MinimumExampleCountPerLeaf: 30
SubsampleFraction: 0.025
FeatureFraction: 0.980
L2Regularization: 0.500

先月中旬頃の自分の記事確認してたら、こんな感じが良いモデル出てるって事で改めて5回程試すと確かにRMSE 1.5910～1.5931でした(笑) これでは4回目が一番だった。

2,427R	1点	芝(1,149R)	ダート(1,189R)	障害(89R)	8頭以下(129R)	9～12頭(597R)	13頭以上(1,701R)	多点
単勝	19.90% (69.19%)	19.15% (71.42%)	20.94% (66.73%)	15.73% (73.26%)	27.13% (69.38%)	21.61% (71.07%)	18.75% (68.51%)	48.62% (72.17%)
複勝	49.44% (78.79%)	45.43% (75.31%)	53.32% (82.72%)	49.44% (71.24%)	59.69% (86.82%)	52.60% (80.44%)	47.56% (77.61%)	85.04% (78.90%)
枠連	11.31% (78.43%)	10.29% (94.24%)	12.38% (66.64%)	8.86% (48.86%)	-- (--)	10.55% (59.58%)	10.93% (80.57%)	27.53% (83.22%)
馬連	7.91% (67.61%)	7.22% (76.01%)	8.75% (61.81%)	5.62% (36.85%)	15.50% (77.29%)	7.71% (61.73%)	7.41% (68.95%)	19.61% (70.77%)
ワイド	21.43% (81.88%)	18.10% (75.92%)	24.73% (88.39%)	20.22% (71.80%)	31.78% (65.35%)	23.62% (81.16%)	19.87% (83.39%)	39.93% (75.69%)
馬単	4.12% (60.97%)	3.57% (74.06%)	4.79% (51.49%)	2.25% (18.65%)	8.53% (78.68%)	4.86% (55.14%)	3.53% (61.67%)	19.61% (68.60%)
三連複	5.23% (86.02%)	3.92% (96.59%)	6.48% (78.85%)	5.62% (45.39%)	8.53% (44.73%)	6.03% (75.01%)	4.70% (93.02%)	13.84% (76.44%)
三連単	0.95% (84.14%)	0.87% (149.14%)	1.01% (26.02%)	1.12% (21.35%)	3.10% (113.80%)	1.17% (64.24%)	0.71% (88.87%)	13.84% (71.43%)
総合	50.97% (75.85%)	47.26% (88.99%)	54.67% (65.32%)	49.44% (48.42%)	62.02% (76.58%)	54.10% (68.54%)	49.03% (77.82%)	85.37% (72.84%)

AIに言わせると、この辺りが純粋に走破タイムを予測させる限界だからと何度も言われ続け、いやね、既にその為の対処として相対タイムやランキングもためしてますって(笑) ただ、新馬戦用走破タイム学習はもう少し良い感じになってるから諦めきれてないんだよね。

SubsampleFractionを0.025→0.050にして5回試すとRMSE 1.5909～1.5932だった。

2,427R	1点	芝(1,149R)	ダート(1,189R)	障害(89R)	8頭以下(129R)	9～12頭(597R)	13頭以上(1,701R)	多点
単勝	19.78% (68.64%)	19.76% (73.91%)	20.19% (64.75%)	14.61% (52.58%)	27.91% (70.70%)	22.11% (66.65%)	18.34% (69.18%)	48.95% (76.39%)
複勝	49.65% (79.57%)	46.13% (78.97%)	52.73% (79.98%)	53.93% (81.80%)	63.57% (93.80%)	52.26% (76.67%)	47.68% (79.51%)	85.00% (80.47%)
枠連	11.72% (74.64%)	10.90% (83.85%)	12.64% (68.21%)	8.86% (51.27%)	-- (--)	10.55% (55.73%)	11.46% (77.03%)	27.31% (81.11%)
馬連	8.16% (76.46%)	7.40% (80.46%)	9.00% (75.14%)	6.74% (42.36%)	17.05% (77.52%)	7.54% (60.32%)	7.70% (82.04%)	19.78% (71.76%)
ワイド	21.63% (89.17%)	18.62% (84.53%)	24.56% (93.92%)	21.35% (85.62%)	34.88% (68.53%)	22.95% (80.85%)	20.16% (93.65%)	41.08% (81.10%)
馬単	4.33% (69.63%)	4.18% (79.96%)	4.63% (62.13%)	2.25% (36.63%)	7.75% (51.55%)	5.36% (72.13%)	3.70% (70.13%)	19.78% (70.25%)
三連複	5.27% (75.96%)	3.74% (76.70%)	6.73% (72.09%)	5.62% (117.98%)	8.53% (34.65%)	6.20% (58.73%)	4.70% (85.14%)	13.97% (77.80%)
三連単	0.78% (31.33%)	0.52% (31.98%)	1.01% (30.77%)	1.12% (30.67%)	0.78% (9.69%)	1.51% (74.77%)	0.53% (17.73%)	13.97% (77.01%)
総合	51.30% (70.63%)	48.30% (73.62%)	53.99% (68.37%)	53.93% (62.52%)	66.67% (58.06%)	53.43% (68.23%)	49.38% (71.80%)	85.37% (76.56%)

RMSEの差はたったの0.0001なんだが、随分と違う感じだ。ひとつひとつでは判断難しいけど、総合的には的中率は向上してるのはその差かな?

迷走中

ズルズルと何度も繰り返して色々試すけど納得出来るモデルが出ない😞

パラメータ / 指標	条件 1	条件 2	条件 3	条件 4	条件 5	条件 6	条件 7
NumberOfLeaves	79	63	63	63	63	63	63
NumberOfIterations	5,000	3,000	3,000	3,000	3,000	3,000	3,000
LearningRate	0.010	0.010	0.010	0.010	0.010	0.010	0.010
MaximumBin	127	63	63	63	63	63	63
MinimumExample	20	20	20	30	40	50	60
SubsampleFraction	0.050	0.050	0.025	0.050	0.050	0.050	0.050
FeatureFraction	0.800	0.800	0.980	0.800	0.800	0.800	0.800
L2Regularization	0.500	0.500	0.500	0.500	0.500	0.500	0.500
R² （決定係数）	0.2263	0.2334	0.2323	0.2340	0.2340	0.2339	0.2334

まあ、葉の最小は30か40みたいなので30に絞って10回試す。でも、これってAIに毎回質問したりしてるけど、AIも所詮ネット上の情報なんかを元にしてるだけで、実際に全てのパターンを試してる訳じゃないので途方もないパターンの中から最適な設定を探すのは地道なパターン潰ししかない気はしてるんだよね。いや、まあ、それが面倒だからAutoMLが割る訳なんだが(笑)

もしかしたら、またAutoMLで24時間とか回してそのパラメーターを真似するのが最短なのかもしれないけど^^; 新馬戦用走破タイム学習だと葉の数が滅茶苦茶多かったんだよね。

30を10回やったが0.2330～0.2337しか出なかったorz なので40を10回スタートした。まあ、以前は40回とかやってた時だいたい20回程度で良いのが出た事もあったし、これ本当に運任せだから辛い😩なんだろなぁ、0.2334～0.2339しか出ない。でまあ、50を10回やったんだが、ここで、0.2332～0.2341でした😝

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	23.91% (78.26%)	23.94% (82.92%)	24.22% (74.10%)	19.10% (69.44%)	33.76% (71.53%)	25.11% (68.42%)	22.60% (82.51%)	53.45% (76.84%)
複勝	54.87% (80.56%)	53.13% (79.22%)	56.40% (81.63%)	58.43% (84.72%)	63.06% (83.50%)	59.10% (80.17%)	52.57% (80.44%)	88.12% (80.68%)
枠連	13.03% (71.94%)	11.95% (68.62%)	13.89% (74.42%)	15.19% (80.51%)	-- (--)	13.98% (70.39%)	11.91% (68.29%)	28.66% (72.86%)
馬連	9.66% (66.23%)	9.15% (65.15%)	10.06% (66.99%)	11.24% (71.24%)	20.38% (97.20%)	11.58% (70.17%)	8.00% (62.05%)	22.76% (69.48%)
ワイド	23.68% (76.26%)	22.38% (71.77%)	25.26% (82.08%)	20.22% (59.44%)	37.58% (74.78%)	27.37% (77.19%)	21.09% (76.04%)	45.63% (76.79%)
馬単	5.52% (70.54%)	5.16% (73.52%)	5.87% (67.51%)	5.62% (70.00%)	12.74% (88.85%)	6.77% (71.50%)	4.42% (68.57%)	22.76% (69.10%)
三連複	6.32% (74.61%)	6.26% (70.16%)	6.44% (77.38%)	5.62% (99.78%)	13.38% (95.22%)	6.77% (76.32%)	5.54% (72.16%)	16.05% (71.69%)
三連単	1.42% (63.95%)	1.17% (42.94%)	1.69% (87.92%)	1.12% (30.67%)	2.55% (51.46%)	2.26% (87.28%)	1.01% (56.36%)	16.05% (65.71%)
総合	56.13% (72.80%)	54.38% (69.30%)	57.60% (76.52%)	60.67% (70.58%)	65.61% (80.36%)	59.85% (75.18%)	53.91% (70.80%)	88.35% (69.53%)

でも、まだまだなモデルでしたorz

追記 2026.10.3 23:39

NumberOfLeaves: 31
NumberOfIterations: 3,000
LearningRate: 0.010
MaximumBinCountPerFeature: 63
MinimumExampleCountPerLeaf: 50
SubsampleFraction: 0.050
FeatureFraction: 0.800
L2Regularization: 0.500

葉の数を更に絞ってみたらR² 0.2371～0.2375だったので0.2375検証すると

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	24.02% (76.18%)	24.26% (79.26%)	24.14% (74.30%)	19.10% (57.98%)	32.48% (71.08%)	26.17% (71.20%)	22.48% (78.47%)	53.03% (76.37%)
複勝	55.29% (80.56%)	54.54% (81.55%)	56.23% (80.04%)	52.81% (73.60%)	64.97% (91.08%)	60.45% (81.31%)	52.52% (79.36%)	88.24% (81.03%)
枠連	13.46% (75.75%)	12.41% (72.37%)	14.39% (78.95%)	13.92% (74.30%)	-- (--)	14.44% (71.68%)	12.30% (72.81%)	29.30% (77.29%)
馬連	9.77% (65.36%)	9.47% (67.28%)	10.14% (64.05%)	8.99% (56.18%)	20.38% (97.20%)	11.73% (70.99%)	8.11% (60.47%)	22.91% (74.36%)
ワイド	23.87% (75.13%)	22.38% (71.10%)	25.66% (79.90%)	20.22% (66.52%)	37.58% (72.99%)	27.97% (79.94%)	21.14% (73.53%)	45.36% (78.11%)
馬単	5.17% (59.82%)	5.01% (69.34%)	5.31% (49.77%)	5.62% (63.60%)	10.19% (58.03%)	6.92% (70.32%)	4.08% (56.08%)	22.91% (73.43%)
三連複	6.55% (82.69%)	6.42% (74.92%)	6.76% (89.44%)	5.62% (99.78%)	13.38% (91.02%)	7.22% (87.53%)	5.70% (80.15%)	16.40% (73.56%)
三連単	1.38% (68.10%)	1.25% (59.13%)	1.53% (79.99%)	1.12% (30.67%)	2.55% (51.46%)	2.26% (119.47%)	0.95% (50.45%)	16.40% (67.78%)
総合	56.70% (72.91%)	55.87% (71.86%)	57.68% (74.53%)	55.06% (65.20%)	66.88% (76.12%)	61.35% (81.56%)	54.08% (68.91%)	88.47% (71.85%)

少しは良くなった。まだまだ改善の余地あるか?

相対タイム学習

ここ最近のパラメーターで相対タイム学習に挑んでますが、

NumberOfLeaves: 31
NumberOfIterations: 1,500
LearningRate: 0.050
MaximumBinCountPerFeature: 127
MinimumExampleCountPerLeaf: 20
SubsampleFraction: 0.500
FeatureFraction: 0.980
L2Regularization: 0.500

これで10回学習させて\(R^{2}\) 0.2288だと

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	23.60% (76.21%)	23.71% (79.50%)	23.89% (74.22%)	17.98% (56.74%)	31.85% (67.39%)	26.02% (69.91%)	21.98% (79.32%)	52.80% (76.59%)
複勝	55.02% (81.02%)	53.91% (80.20%)	56.32% (82.34%)	52.81% (74.16%)	64.33% (86.69%)	60.60% (82.24%)	52.13% (80.06%)	88.35% (80.94%)
枠連	12.48% (68.05%)	11.86% (64.13%)	13.12% (72.68%)	11.39% (52.91%)	-- (--)	13.83% (69.47%)	11.24% (63.53%)	28.58% (75.07%)
馬連	9.27% (60.49%)	9.15% (60.52%)	9.49% (61.49%)	7.87% (46.29%)	20.38% (99.04%)	11.28% (69.32%)	7.55% (53.83%)	22.22% (73.03%)
ワイド	23.26% (72.21%)	22.22% (66.60%)	24.54% (78.30%)	20.22% (67.53%)	37.58% (70.06%)	27.97% (80.86%)	20.25% (69.18%)	45.06% (76.93%)
馬単	4.83% (53.25%)	4.46% (55.09%)	5.23% (52.27%)	4.49% (40.45%)	10.83% (88.22%)	6.32% (57.53%)	3.75% (48.59%)	22.22% (70.93%)
三連複	6.25% (76.38%)	5.95% (82.28%)	6.60% (72.81%)	5.62% (41.69%)	11.46% (64.90%)	6.62% (62.47%)	5.65% (82.57%)	15.36% (69.27%)
三連単	1.46% (93.93%)	1.10% (109.68%)	1.77% (78.67%)	2.25% (80.90%)	2.55% (102.04%)	1.80% (50.11%)	1.23% (109.52%)	15.36% (64.91%)
総合	56.21% (72.75%)	54.77% (74.95%)	57.76% (71.59%)	55.06% (57.65%)	66.24% (82.62%)	61.50% (67.74%)	53.36% (73.32%)	88.54% (69.50%)

これを更に精度上げるにはってGeminiにぶつけると、

NumberOfLeaves: 63
NumberOfIterations: 3,000
LearningRate: 0.010
MaximumBinCountPerFeature: 127
MinimumExampleCountPerLeaf: 20
SubsampleFraction: 0.500
FeatureFraction: 0.800
L2Regularization: 0.500

これを試すと\(R^{2}\) 0.2339

2,610R	1点	芝(1,278R)	ダート(1,243R)	障害(89R)	8頭以下(157R)	9～12頭(665R)	13頭以上(1,788R)	多点
単勝	23.56% (77.59%)	23.47% (82.89%)	24.05% (73.68%)	17.98% (56.07%)	31.85% (68.22%)	25.56% (68.45%)	22.09% (81.81%)	53.75% (78.08%)
複勝	55.13% (80.39%)	53.99% (80.43%)	56.07% (80.26%)	58.43% (81.80%)	63.69% (84.08%)	60.00% (80.48%)	52.57% (80.04%)	88.28% (80.70%)
枠連	13.03% (71.22%)	11.76% (67.45%)	14.23% (75.50%)	12.66% (59.11%)	-- (--)	13.53% (68.15%)	12.08% (68.18%)	29.17% (75.76%)
馬連	9.50% (65.75%)	9.08% (64.98%)	9.98% (67.47%)	8.99% (52.81%)	19.11% (99.75%)	11.13% (66.66%)	8.05% (62.43%)	23.45% (72.81%)
ワイド	23.64% (74.45%)	22.07% (65.40%)	25.34% (84.42%)	22.47% (65.28%)	36.94% (70.96%)	27.07% (75.61%)	21.20% (74.33%)	45.79% (76.73%)
馬単	5.25% (69.81%)	5.09% (79.67%)	5.47% (61.31%)	4.49% (46.85%)	10.83% (92.61%)	6.62% (64.06%)	4.25% (69.94%)	23.45% (72.62%)
三連複	6.32% (75.02%)	6.03% (66.40%)	6.76% (87.28%)	4.49% (27.64%)	12.74% (66.82%)	6.62% (72.09%)	5.65% (76.83%)	15.94% (73.48%)
三連単	1.42% (56.84%)	1.41% (58.73%)	1.45% (56.76%)	1.12% (30.67%)	2.55% (105.73%)	2.26% (73.25%)	1.01% (46.44%)	15.94% (67.97%)
総合	56.36% (71.39%)	55.16% (70.81%)	57.28% (73.32%)	60.67% (52.44%)	66.24% (84.02%)	60.60% (71.09%)	53.91% (70.00%)	88.51% (71.66%)

なんだけど、微妙な検証結果。