Adaboost算法实现一

来源：互联网发布：炸奥利奥知乎编辑：程序博客网时间：2024/06/06 11:47

对前四次迭代分析

训练数据

实验样本：30个

43.8744 84.0717 1

38.1558 25.4282 2

76.5517 81.4285 1

79.5200 24.3525 2

18.6873 92.9264 1

48.9764 34.9984 2

44.5586 19.6595 2

64.6313 25.1084 2

70.9365 61.6045 2

75.4687 47.3289 2

27.6025 35.1660 1

67.9703 83.0829 1

65.5098 58.5264 2

16.2612 54.9724 1

11.8998 91.7194 1

49.8364 28.5839 2

95.9744 75.7200 2

34.0386 75.3729 1

58.5268 38.0446 2

22.3812 56.7822 1

75.1267 7.5854 2

25.5095 5.3950 2

50.5957 53.0798 1

69.9077 77.9167 1

89.0903 93.4011 1

95.9291 12.9906 2

54.7216 56.8824 1

13.8624 46.9391 1

14.9294 1.1902 2

25.7508 33.7123 1

第一次循环

adaboost_model.parameters{1}

ans =

min_error: 0.2000

min_error_thr: 31.1481

pos_neg: 'neg'

dim: 2

mm:10

[L,hits,error_rate] = threshold_te(adaboost_model.parameters{turn},...

tr_set,samples_weight,tr_labels)

L =

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

hits =

error_rate =

0.2000

第一次分类以后权重的分配

samples_weight(t_labeled) = samples_weight(t_labeled)*...%提取正确分类的权重

((error_rate)/(1-error_rate))

将错误的分类突出出来，标记的权重更大

samples_weight =

0.0083

0.0333

0.0083

0.0333

0.0083

0.0333

0.0083

0.0333

0.0083

0.0333

0.0083

第二次循环

adaboost_model.parameters{2}

ans =

min_error: 0.1667

min_error_thr: 76.8184

pos_neg: 'neg'

dim: 2

mm:24

[L,hits,error_rate] = threshold_te(adaboost_model.parameters{turn},...

tr_set,samples_weight,tr_labels)

L =

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

hits =

error_rate =

0.1667

权重

weights: [0.6021 0.6990]

第二次更新样本权重，没归一化之前的

samples_weight(t_labeled) = samples_weight(t_labeled)*...%提取正确分类的权重

((error_rate)/(1-error_rate))

samples_weight =

0.0042

0.0167

0.0042

0.0167

0.0208

0.0042

0.0167

0.0208

0.0042

0.0167

0.0208

0.0167

0.0208

0.0042

0.0208

0.0042

0.0208

0.0042

0.0208

第三次迭代

adaboost_model.parameters{3}

ans =

min_error: 0.1625

min_error_thr: 56.6242

pos_neg: 'pos'

dim: 1

mm:24

[L,hits,error_rate] = threshold_te(adaboost_model.parameters{turn},...

tr_set,samples_weight,tr_labels)

L =

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

hits =

error_rate =

0.1625

weights: [0.6021 0.6990 0.7121]

samples_weight =

0.0024

0.0125

0.0024

0.0500

0.0125

0.0024

0.0097

0.0121

0.0125

0.0097

0.0121

0.0024

0.0125

0.0097

0.0121

0.0097

0.0121

0.0024

0.0125

0.0121

0.0125

0.0024

0.0121

0.0125

0.0121

第四次迭代

adaboost_model.parameters{4}

ans =

min_error: 0.1866

min_error_thr: 35.0822

pos_neg: 'neg'

dim: 2

mm:12

[L,hits,error_rate] = threshold_te(adaboost_model.parameters{turn},...

tr_set,samples_weight,tr_labels)

L =

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

1 0

0 1

hits =

error_rate =

0.1866

weights: [0.6021 0.6990 0.7121 0.6395]

samples_weight(t_labeled) = samples_weight(t_labeled)*...%提取正确分类的权重

((error_rate)/(1-error_rate))

samples_weight =

0.0017

0.0088

0.0017

0.0353

0.0088

0.0017

0.0299

0.0086

0.0088

0.0299

0.0086

0.0017

0.0088

0.0299

0.0086

0.0299

0.0086

0.0017

0.0088

0.0086

0.0088

0.0017

0.0086

0.0088

0.0373