交叉验证(Cross-validation)主要用于建模应用中,例如PCR 、PLS 回归建模中。在给定的建模样本中,拿出大部分样本进行建模型,留小部分样本用刚建立的模型进行预报,并求这小部分样本的预报误差,记录它们的平方加和。这个过程一直进行,直到所有的样本都被预报了一次而且仅被预报一次。把每个样本的预报误差平方加和,称为PRESS(predicted Error Sum of Squares)。交叉验证方法在克服过拟合(Over-Fitting)问题上非常有用。
K-fold cross-validation
{{K折交叉验证,初始采样分割成K个子样本,一个单独的子样本被保留作为验证模型的数据,其他K-1个样本用来训练。交叉验证重复K次,每个子样本验证一次,平均K次的结果或者使用其它结合方式,最终得到一个单一估测。这个方法的优势在于,同时重复运用随机产生的子样本进行训练和验证,每次的结果验证一次,10折交叉验证是最常用的。}}
CVlm {DAAG}
val=CVlm(df=cv,m=10,form.lm=formula(Y~X1+X2+X3+X4))# m=10(10-fold,df=cv为数据框文件为cv,拟和普通最小二乘法)
Analysis of Variance Table Response: Y
Df Sum Sq Mean Sq F value Pr(>F)
X1 1 69.4 69.4 17.19 0.00042
X2 1 4.1 4.1 1.03 0.32210
X3 1 32.3 32.3 8.01 0.00974
X4 1 27.8 27.8 6.88 0.01552
Residuals 22 88.8 4.0
X1 ***
X2
X3 **
X4 *
Residuals
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
fold 1
Observations in test set: 2
13 16
Predicted 12.03 10.180
cvpred 13.49 10.768
Y 8.40 10.100
CV residual -5.09 -0.668
Sum of squares = 26.4 Mean square = 13.2 n = 2
fold 2
Observations in test set: 3
8 19 26
Predicted 13.52 12.03 8.85
cvpred 13.67 12.02 7.78
Y 12.10 10.80 13.30
CV residual -1.57 -1.22 5.52
Sum of squares = 34.4 Mean square = 11.5 n = 3
fold 3
Observations in test set: 3
9 22 25
Predicted 7.87 13.16 17.79
cvpred 8.09 13.22 15.15
Y 9.60 14.90 20.00
CV residual 1.51 1.68 4.85
Sum of squares = 28.7 Mean square = 9.56 n = 3
fold 4
Observations in test set: 3
1 20 27
Predicted 11.428 12.3 11.29
cvpred 11.571 12.5 11.52
Y 11.200 10.2 10.40
CV residual -0.371 -2.3 -1.12
Sum of squares = 6.71 Mean square = 2.24 n = 3
fold 5
Observations in test set: 3
5 17 18
Predicted 11.10 13.05 9.167
cvpred 10.73 12.89 9.229
Y 13.40 14.80 9.100
CV residual 2.67 1.91 -0.129
Sum of squares = 10.8 Mean square = 3.59 n = 3
fold 6
Observations in test set: 3
6 10 21
Predicted 15.33 9.58 12.25
cvpred 13.63 9.76 12.27
Y 18.30 8.40 13.60
CV residual 4.67 -1.36 1.33
Sum of squares = 25.4 Mean square = 8.48 n = 3
fold 7
Observations in test set: 3
12 23 24
Predicted 10.436 15.963 15.21
cvpred 10.486 16.445 15.81
Y 10.600 16.000 13.20
CV residual 0.114 -0.445 -2.61
Sum of squares = 7.03 Mean square = 2.34 n = 3
fold 8
Observations in test set: 3
2 3 11
Predicted 9.48 13.064 11.87
cvpred 9.91 13.202 12.32
Y 8.80 12.300 9.30
CV residual -1.11 -0.902 -3.02
Sum of squares = 11.2 Mean square = 3.72 n = 3
fold 9
Observations in test set: 2
4 7
Predicted 10.716 11.64
cvpred 10.646 12.21
Y 11.600 11.10
CV residual 0.954 -1.11
Sum of squares = 2.13 Mean square = 1.07 n = 2
fold 10
Observations in test set: 2
14 15
Predicted 11.26 11.441
cvpred 11.75 11.373
Y 9.60 10.900
CV residual -2.15 -0.473
Sum of squares = 4.84 Mean square = 2.42 n = 2
Overall (Sum over all 2 folds)
ms 5.83 #10折平均的均方为5.83