10折交叉验证深入理解

r囧r小猫 2022-04-02 05:41 474阅读 0赞

交叉验证（Cross Validation），有的时候也称作循环估计（Rotation Estimation），是一种统计学上将数据样本切割成较小子集的实用方法，该理论是由Seymour Geisser提出的。  
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)。  
常用的精度测试方法主要是交叉验证，例如10折交叉验证(10-fold cross validation)，将数据集分成十份，轮流将其中9份做训练1份做验证，10次的结果的均值作为对算法精度的估计，一般还需要进行**多次10折**交叉验证**求均值**，例如：10次10折交叉验证，以求更精确一点。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。

"""这里是正确的代码"""
    def cross_10folds(data,folds,jiange,start_index,end_index):
        df_test=data[start_index*jiange:end_index*jiange]  #数据刚好可以做10折交叉验证。
        df_test_index=list(df_test.index)
        df_test_flag=data.index.isin(df_test_index)  #都转换为list来判定成员资格
        diff_flag = [not f for f in df_test_flag] # 不是df_test_flag里面的索引的索引记为df_train_index的索引集合
        df_train= data[diff_flag]
        return df_train,df_test
    
    path='C:/Users/Administrator/Desktop/zhou1.csv'
    columns1=['cid','side_effects']
    import pandas as pd
    #data=pd.read_csv(path)[:100]  #经验证划分完全正确
    data=pd.read_csv(path)
    data.columns=columns1
    folds=10
    jiange=int(data.shape[0]/folds)
    #10次10折交叉验证的代码
    for i in range(1,11):
    	#将数据集随机打散
    	data=data.sample(frac = 1) #随机打乱样本
    	for i in range(1,folds+1):
    	    df_train,df_test=cross_10folds(data,folds,jiange,i-1,i) #起始和尾部索引
    	    df_train.to_csv('C:/Users/Administrator/Desktop/zhou/train_'+str(i)+'.csv',index=True,header=True) #检验对了再改成False
    	    df_test.to_csv('C:/Users/Administrator/Desktop/zhou/test_'+str(i)+'.csv',index=True,header=True)

1.matlab版本代码实现：

%十折交叉验证\
    [m,n]=size(train_bags);     %蛋白质的数量的个数
    indices=crossvalind('Kfold',m,10);
    temp_Hamming_Loss=zeros(1,10);
    temp_Macro_F1=zeros(1,10);
    temp_Micro_F1=zeros(1,10);
    for k=1:1:10
        test=(indices==k);  %获取验证集，但是最后替换掉了
        train=~test;
        train_data=train_bags(train,:);   %训练集的包
        train_target=train_lable(:,train);  %训练集的标记
        test_bags=train_bags(test,:);          %十折交叉验证的测试集包
        test_target=train_lable(:,test);        %十折交叉验证的测试集标记

2.Python3版本代码实现：  
2.1.简易5折交叉验证分割结果打印:未随机打散的情形下

#简易5折交叉验证分割结果打印:未随机打散的情形下
    from sklearn.cross_validation import KFold
    kf = KFold(150,n_folds=5, shuffle=False)
    for eachfold in kf:
        print(eachfold,'=========') #kf是一个长度为5的list,list中的每个元素是一次10折交叉验证的划分结果。

5折交叉验证划分结果的前半部分是trainset，后半部分是testset

(array([ 30,  31,  32,  33,  34,  35,  36,  37,  38,  39,  40,  41,  42,
            43,  44,  45,  46,  47,  48,  49,  50,  51,  52,  53,  54,  55,
            56,  57,  58,  59,  60,  61,  62,  63,  64,  65,  66,  67,  68,
            69,  70,  71,  72,  73,  74,  75,  76,  77,  78,  79,  80,  81,
            82,  83,  84,  85,  86,  87,  88,  89,  90,  91,  92,  93,  94,
            95,  96,  97,  98,  99, 100, 101, 102, 103, 104, 105, 106, 107,
           108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120,
           121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133,
           134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146,
           147, 148, 149]), array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
           17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29]))
    ====================================================
    (array([  0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,
            13,  14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25,
            26,  27,  28,  29,  60,  61,  62,  63,  64,  65,  66,  67,  68,
            69,  70,  71,  72,  73,  74,  75,  76,  77,  78,  79,  80,  81,
            82,  83,  84,  85,  86,  87,  88,  89,  90,  91,  92,  93,  94,
            95,  96,  97,  98,  99, 100, 101, 102, 103, 104, 105, 106, 107,
           108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120,
           121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133,
           134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146,
           147, 148, 149]), array([30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46,
           47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59]))
    ====================================================
    (array([  0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,
            13,  14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25,
            26,  27,  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,
            39,  40,  41,  42,  43,  44,  45,  46,  47,  48,  49,  50,  51,
            52,  53,  54,  55,  56,  57,  58,  59,  90,  91,  92,  93,  94,
            95,  96,  97,  98,  99, 100, 101, 102, 103, 104, 105, 106, 107,
           108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120,
           121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133,
           134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146,
           147, 148, 149]), array([60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76,
           77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89]))
    ====================================================
    (array([  0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,
            13,  14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25,
            26,  27,  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,
            39,  40,  41,  42,  43,  44,  45,  46,  47,  48,  49,  50,  51,
            52,  53,  54,  55,  56,  57,  58,  59,  60,  61,  62,  63,  64,
            65,  66,  67,  68,  69,  70,  71,  72,  73,  74,  75,  76,  77,
            78,  79,  80,  81,  82,  83,  84,  85,  86,  87,  88,  89, 120,
           121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133,
           134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146,
           147, 148, 149]), array([ 90,  91,  92,  93,  94,  95,  96,  97,  98,  99, 100, 101, 102,
           103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115,
           116, 117, 118, 119]))
    ====================================================
    (array([  0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,
            13,  14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25,
            26,  27,  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,
            39,  40,  41,  42,  43,  44,  45,  46,  47,  48,  49,  50,  51,
            52,  53,  54,  55,  56,  57,  58,  59,  60,  61,  62,  63,  64,
            65,  66,  67,  68,  69,  70,  71,  72,  73,  74,  75,  76,  77,
            78,  79,  80,  81,  82,  83,  84,  85,  86,  87,  88,  89,  90,
            91,  92,  93,  94,  95,  96,  97,  98,  99, 100, 101, 102, 103,
           104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116,
           117, 118, 119]), array([120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132,
           133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145,
           146, 147, 148, 149]))

2.2.随机打散数据并做10次10折交叉验证

# -*- coding: utf-8 -*-
    import numpy as np
    from sklearn import cross_validation
    from sklearn import datasets
    from sklearn import svm
    iris = datasets.load_iris()
    X=iris.data
    Y=iris.target
    
    def tenFolds(X,Y):
        from sklearn.model_selection import StratifiedKFold
        skf= StratifiedKFold(n_splits=10)
        from sklearn.cross_validation import cross_val_score
        clf = svm.SVC(kernel='linear', C=5)
        zhou=[]
        #shoushou=list(skf.split(X,Y)) #发现每一个都是tuple类型的数据
        for train_index,test_index in skf.split(X,Y):
            print('Train: ',train_index,'Test: ',test_index)
            X_train,X_test=X[train_index],X[test_index]
            Y_train,Y_test=Y[train_index],Y[test_index]
            v10=cross_val_score(clf,X_train,Y_train,cv=5).mean() 
            zhou.append(v10)
        print('均值:',np.mean(zhou)) 
        print('方差:',np.std(zhou)) 
        return zhou,np.mean(zhou)+np.std(zhou)
    
    # X:features  Y:targets  cv:k
    import pandas as pd
    all_valid=[]
    mean_var=[]
    #做10次10折交叉验证结果分析
    for i in range(1,11):
        df1=pd.DataFrame(X)
        df2=pd.DataFrame(Y)
        df3=pd.concat([df1,df2],axis=1)
        df3.columns=['f1','f2','f3','f4','label']
        df3=df3.sample(frac = 1) #随机打乱样本
        df3.index=range(150)
        X1=np.array(df3[['f1','f2','f3','f4']])
        Y1=np.array(df3['label'])
        zhou,vsd= tenFolds(X1,Y1)
        all_valid.append(zhou)
        mean_var.append(vsd)
    np.mean(mean_var)

all\_valid

[0.9777777777777776, 0.962962962962963, 0.962962962962963, 0.9703703703703702, 0.9703703703703702, 0.9703703703703702, 0.9777777777777779, 0.9555555555555555, 0.9555555555555555, 0.9703703703703702]
    [0.9555555555555555, 0.9703703703703704, 0.962962962962963, 0.9555555555555555, 0.9703703703703704, 0.962962962962963, 0.9703703703703702, 0.9925925925925926, 0.9703703703703702, 0.9777777777777776]
    [0.9703703703703704, 0.9777777777777779, 0.9555555555555555, 0.9703703703703704, 0.9703703703703704, 0.9481481481481481, 0.9703703703703704, 0.962962962962963, 0.962962962962963,
     0.9777777777777779]
    [0.9703703703703702, 0.9703703703703702, 0.962962962962963, 0.9851851851851852, 0.962962962962963, 0.962962962962963, 
    0.9703703703703702, 0.9777777777777779, 0.9703703703703702, 0.9703703703703702]
    [0.9777777777777779, 0.962962962962963, 0.9555555555555555, 0.962962962962963, 0.962962962962963, 0.9555555555555555, 
    0.9703703703703702, 0.9777777777777776, 0.9407407407407407, 0.9777777777777776]
    [0.9481481481481481, 0.9703703703703704, 0.9777777777777779, 0.9407407407407409, 0.962962962962963, 0.9555555555555555, 0.9555555555555555, 0.962962962962963, 0.9703703703703702, 0.9851851851851852]
    [0.9777777777777776, 0.9777777777777776, 0.9777777777777776, 0.9703703703703704, 0.9925925925925926, 0.9851851851851852, 0.9777777777777779, 0.9703703703703704, 0.9703703703703704, 0.9851851851851852]
    [0.962962962962963, 0.9703703703703702, 0.9703703703703704, 0.962962962962963, 0.9703703703703702, 0.9777777777777779, 0.9481481481481481, 0.9555555555555555, 0.9481481481481481, 0.9555555555555557]
    [0.962962962962963, 0.9703703703703702, 0.9555555555555555, 0.9703703703703704, 0.9703703703703702, 0.962962962962963, 0.9703703703703702, 0.962962962962963, 0.9555555555555555, 
    0.962962962962963]
    [0.9703703703703702, 0.962962962962963, 0.9555555555555555, 0.962962962962963, 0.9703703703703702, 0.9703703703703702, 0.9777777777777779, 0.9851851851851852, 0.9777777777777779, 0.9703703703703702]

mean\_var

0.9749615103905077
    0.9792592592592592
    0.9755863663546611
    0.9769957569703696
    0.975823919626472
    0.9757929689449547
    0.9855066526904122
    0.9718518518518519
    0.9699876405729984
    0.9784847786297061

分层交叉验证和不平衡数据集的处理：  
[https://blog.csdn.net/asialee\_bird/article/details/83714612][https_blog.csdn.net_asialee_bird_article_details_83714612]  
Python smote例程：使用sklearn的make\_classification生成不平衡数据样本  
[https://blog.csdn.net/levy\_cui/article/details/86707049][https_blog.csdn.net_levy_cui_article_details_86707049]

[https_blog.csdn.net_asialee_bird_article_details_83714612]: https://blog.csdn.net/asialee_bird/article/details/83714612
[https_blog.csdn.net_levy_cui_article_details_86707049]: https://blog.csdn.net/levy_cui/article/details/86707049