10折交叉验证深入理解

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_testpath='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Falsedf_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=1:1:10test=(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.targetdef tenFolds(X,Y):from sklearn.model_selection import StratifiedKFoldskf= StratifiedKFold(n_splits=10)from sklearn.cross_validation import cross_val_scoreclf = 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
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

10折交叉验证深入理解相关推荐

机器学习 - 随机森林手动10 折交叉验证
随机森林的 10 折交叉验证再回到之前的随机森林(希望还没忘记,机器学习算法-随机森林初探(1)) library(randomForest) set.seed(304) rf1000 <- ...
R语言caret包构建xgboost模型实战：特征工程（连续数据离散化、因子化、无用特征删除）、配置模型参数（随机超参数寻优、10折交叉验证）并训练模型
R语言caret包构建xgboost模型实战:特征工程(连续数据离散化.因子化.无用特征删除).配置模型参数(随机超参数寻优.10折交叉验证)并训练模型目录
matlab.10折交叉验证
clc clear all % 导入数据 data = load('F:\work_matlab\Matlab\wdbc.txt'); [data_r, data_c] = size(data); % ...
10折交叉验证（10-fold Cross Validation）与留一法（Leave-One-Out）、分层采样（Stratification）
10折交叉验证我们构建一个分类器,输入为运动员的身高.体重,输出为其从事的体育项目-体操.田径或篮球. 一旦构建了分类器,我们就可能有兴趣回答类似下述的问题: 1. 该分类器的精确率怎么样? 2. 该 ...
《写给程序员的数据挖掘实践指南》——5.2. 10折交叉验证的例子
本节书摘来自异步社区出版社<写给程序员的数据挖掘实践指南>一书中的第5章,第5.2节,作者:[美]Ron Zacharski(扎哈尔斯基),更多章节内容可以访问云栖社区"异步社区 ...
《机器学习》课后习题 3.4 选择两个 UCI 数据集，比较 10 折交叉验证法和留法所估计出的对率回归的错误率.
参考了han同学的答案,数据集也可在han同学的github上下载. 3.4 选择两个 UCI 数据集,比较 10 折交叉验证法和留法所估计出的对率回归的错误率. import numpy as n ...
10折交叉验证（10-fold Cross Validation）与留一法（Leave-One-Out）、分层采样（Stratification）...
10折交叉验证我们构建一个分类器,输入为运动员的身高.体重,输出为其从事的体育项目-体操.田径或篮球. 一旦构建了分类器,我们就可能有兴趣回答类似下述的问题: 1. 该分类器的精确率怎么样? 2. ...
matlab-K折交叉验证与分层K折交叉验证
文章目录 K折交叉验证有什么用? 如何实现K折交叉验证? K折交叉验证的要点:(文字版) 如何实现K折交叉验证(图片版) 如何实现K折交叉验证(matlab版) 为啥我们需要分层K折交叉验证? 如何实 ...
Pytorch最简单的图像分类——K折交叉验证处理小型鸟类数据集分类2.0版本ing
https://blog.csdn.net/hb_learing/article/details/110411532 https://blog.csdn.net/Pl_Sun/article/deta ...
十折交叉验证10-fold cross validation, 数据集划分训练集验证集测试集
机器学习数据挖掘数据集划分训练集验证集测试集 Q:如何将数据集划分为测试数据集和训练数据集? A:three ways: 1.像sklearn一样,提供一个将数据集切分成训练集和测试集的函数 ...

10折交叉验证深入理解

10折交叉验证深入理解相关推荐

最新文章

热门文章