使用sklearn的cross_val_score進行交叉驗證實例

from sklearn import datasets #自帶數(shù)據(jù)集
from sklearn.model_selection import train_test_split,cross_val_score #劃分數(shù)據(jù) 交叉驗證
from sklearn.neighbors import KNeighborsClassifier #一個簡單的模型，只有K一個參數(shù)，類似K-means
import matplotlib.pyplot as plt
iris = datasets.load_iris() #加載sklearn自帶的數(shù)據(jù)集
X = iris.data #這是數(shù)據(jù)
y = iris.target #這是每個數(shù)據(jù)所對應的標簽
train_X,test_X,train_y,test_y = train_test_split(X,y,test_size=1/3,random_state=3) #這里劃分數(shù)據(jù)以1/3的來劃分 訓練集訓練結果 測試集測試結果
k_range = range(1,31)
cv_scores = [] #用來放每個模型的結果值
for n in k_range:
 knn = KNeighborsClassifier(n) #knn模型，這里一個超參數(shù)可以做預測，當多個超參數(shù)時需要使用另一種方法GridSearchCV
 scores = cross_val_score(knn,train_X,train_y,cv=10,scoring='accuracy') #cv：選擇每次測試折數(shù) accuracy：評價指標是準確度,可以省略使用默認值，具體使用參考下面。
 cv_scores.append(scores.mean())
plt.plot(k_range,cv_scores)
plt.xlabel('K')
plt.ylabel('Accuracy') #通過圖像選擇最好的參數(shù)
plt.show()
best_knn = KNeighborsClassifier(n_neighbors=3) # 選擇最優(yōu)的K=3傳入模型
best_knn.fit(train_X,train_y) #訓練模型
print(best_knn.score(test_X,test_y)) #看看評分

import os
import numpy as np
import pandas as pd
from sklearn import datasets
from sklearn import preprocessing
from sklearn import neighbors
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn import svm
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.model_selection import StratifiedKFold
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
from time import time
from sklearn.naive_bayes import MultinomialNB
from sklearn import tree
from sklearn.ensemble import GradientBoostingClassifier

#讀取sklearn自帶的數(shù)據(jù)集（鳶尾花）
def getData_1():
 iris = datasets.load_iris()
 X = iris.data #樣本特征矩陣，150*4矩陣，每行一個樣本，每個樣本維度是4
 y = iris.target #樣本類別矩陣，150維行向量，每個元素代表一個樣本的類別
#讀取本地excel表格內(nèi)的數(shù)據(jù)集（抽取每類60%樣本組成訓練集，剩余樣本組成測試集）
#返回一個元祖，其內(nèi)有4個元素（類型均為numpy.ndarray）：
#（1）歸一化后的訓練集矩陣，每行為一個訓練樣本，矩陣行數(shù)=訓練樣本總數(shù)，矩陣列數(shù)=每個訓練樣本的特征數(shù)
#（2）每個訓練樣本的類標
#（3）歸一化后的測試集矩陣，每行為一個測試樣本，矩陣行數(shù)=測試樣本總數(shù)，矩陣列數(shù)=每個測試樣本的特征數(shù)
#（4）每個測試樣本的類標
#【注】歸一化采用“最大最小值”方法。
def getData_2():
 fPath = 'D:\分類算法\binary_classify_data.txt'
 if os.path.exists(fPath):
  data = pd.read_csv(fPath,header=None,skiprows=1,names=['class0','pixel0','pixel1','pixel2','pixel3'])
  X_train1, X_test1, y_train1, y_test1 = train_test_split(data, data['class0'], test_size = 0.4, random_state = 0)
  min_max_scaler = preprocessing.MinMaxScaler() #歸一化
  X_train_minmax = min_max_scaler.fit_transform(np.array(X_train1))
  X_test_minmax = min_max_scaler.fit_transform(np.array(X_test1))
  return (X_train_minmax, np.array(y_train1), X_test_minmax, np.array(y_test1))
 else:
  print ('No such file or directory!')

#讀取本地excel表格內(nèi)的數(shù)據(jù)集（每類隨機生成K個訓練集和測試集的組合）
#【K的含義】假設一共有1000個樣本，K取10，那么就將這1000個樣本切分10份（一份100個），那么就產(chǎn)生了10個測試集
#對于每一份的測試集，剩余900個樣本即作為訓練集
#結果返回一個字典：鍵為集合編號（1train, 1trainclass, 1test, 1testclass, 2train, 2trainclass, 2test, 2testclass...），值為數(shù)據(jù)
#其中1train和1test為隨機生成的第一組訓練集和測試集（1trainclass和1testclass為訓練樣本類別和測試樣本類別），其他以此類推
def getData_3():
 fPath = 'D:\\分類算法\\binary_classify_data.txt'
 if os.path.exists(fPath):
  #讀取csv文件內(nèi)的數(shù)據(jù)，
  dataMatrix = np.array(pd.read_csv(fPath,header=None,skiprows=1,names=['class0','pixel0','pixel1','pixel2','pixel3']))
  #獲取每個樣本的特征以及類標
  rowNum, colNum = dataMatrix.shape[0], dataMatrix.shape[1]
  sampleData = []
  sampleClass = []
  for i in range(0, rowNum):
   tempList = list(dataMatrix[i,:])
   sampleClass.append(tempList[0])
   sampleData.append(tempList[1:])
  sampleM = np.array(sampleData) #二維矩陣，一行是一個樣本，行數(shù)=樣本總數(shù)，列數(shù)=樣本特征數(shù)
  classM = np.array(sampleClass) #一維列向量，每個元素對應每個樣本所屬類別
  #調用StratifiedKFold方法生成訓練集和測試集
  skf = StratifiedKFold(n_splits = 10)
  setDict = {} #創(chuàng)建字典，用于存儲生成的訓練集和測試集
  count = 1
  for trainI, testI in skf.split(sampleM, classM):
   trainSTemp = [] #用于存儲當前循環(huán)抽取出的訓練樣本數(shù)據(jù)
   trainCTemp = [] #用于存儲當前循環(huán)抽取出的訓練樣本類標
   testSTemp = [] #用于存儲當前循環(huán)抽取出的測試樣本數(shù)據(jù)
   testCTemp = [] #用于存儲當前循環(huán)抽取出的測試樣本類標
   #生成訓練集
   trainIndex = list(trainI)
   for t1 in range(0, len(trainIndex)):
    trainNum = trainIndex[t1]
    trainSTemp.append(list(sampleM[trainNum, :]))
    trainCTemp.append(list(classM)[trainNum])
   setDict[str(count) + 'train'] = np.array(trainSTemp)
   setDict[str(count) + 'trainclass'] = np.array(trainCTemp)
   #生成測試集
   testIndex = list(testI)
   for t2 in range(0, len(testIndex)):
    testNum = testIndex[t2]
    testSTemp.append(list(sampleM[testNum, :]))
    testCTemp.append(list(classM)[testNum])
   setDict[str(count) + 'test'] = np.array(testSTemp)
   setDict[str(count) + 'testclass'] = np.array(testCTemp)
   count += 1
  return setDict
 else:
  print ('No such file or directory!')
#K近鄰（K Nearest Neighbor）
def KNN():
 clf = neighbors.KNeighborsClassifier()
 return clf

#線性鑒別分析（Linear Discriminant Analysis）
def LDA():
 clf = LinearDiscriminantAnalysis()
 return clf

#支持向量機（Support Vector Machine）
def SVM():
 clf = svm.SVC()
 return clf

#邏輯回歸（Logistic Regression）
def LR():
 clf = LogisticRegression()
 return clf

#隨機森林決策樹（Random Forest）
def RF():
 clf = RandomForestClassifier()
 return clf

#多項式樸素貝葉斯分類器
def native_bayes_classifier():
 clf = MultinomialNB(alpha = 0.01)
 return clf

#決策樹
def decision_tree_classifier():
 clf = tree.DecisionTreeClassifier()
 return clf

#GBDT
def gradient_boosting_classifier():
 clf = GradientBoostingClassifier(n_estimators = 200)
 return clf

#計算識別率
def getRecognitionRate(testPre, testClass):
 testNum = len(testPre)
 rightNum = 0
 for i in range(0, testNum):
  if testClass[i] == testPre[i]:
   rightNum += 1
 return float(rightNum) / float(testNum)

#report函數(shù)，將調參的詳細結果存儲到本地F盤（路徑可自行修改，其中n_top是指定輸出前多少個最優(yōu)參數(shù)組合以及該組合的模型得分）
def report(results, n_top=5488):
 f = open('F:/grid_search_rf.txt', 'w')
 for i in range(1, n_top + 1):
  candidates = np.flatnonzero(results['rank_test_score'] == i)
  for candidate in candidates:
   f.write("Model with rank: {0}".format(i) + '\n')
   f.write("Mean validation score: {0:.3f} (std: {1:.3f})".format(
     results['mean_test_score'][candidate],
     results['std_test_score'][candidate]) + '\n')
   f.write("Parameters: {0}".format(results['params'][candidate]) + '\n')
   f.write("\n")
 f.close()

#自動調參（以隨機森林為例）
def selectRFParam():
 clf_RF = RF()
 param_grid = {"max_depth": [3,15],
     "min_samples_split": [3, 5, 10],
     "min_samples_leaf": [3, 5, 10],
     "bootstrap": [True, False],
     "criterion": ["gini", "entropy"],
     "n_estimators": range(10,50,10)}
     # "class_weight": [{0:1,1:13.24503311,2:1.315789474,3:12.42236025,4:8.163265306,5:31.25,6:4.77326969,7:19.41747573}],
     # "max_features": range(3,10),
     # "warm_start": [True, False],
     # "oob_score": [True, False],
     # "verbose": [True, False]}
 grid_search = GridSearchCV(clf_RF, param_grid=param_grid, n_jobs=4)
 start = time()
 T = getData_2() #獲取數(shù)據(jù)集
 grid_search.fit(T[0], T[1]) #傳入訓練集矩陣和訓練樣本類標
 print("GridSearchCV took %.2f seconds for %d candidate parameter settings."
   % (time() - start, len(grid_search.cv_results_['params'])))
 report(grid_search.cv_results_)

#“主”函數(shù)1（KFold方法生成K個訓練集和測試集，即數(shù)據(jù)集采用getData_3()函數(shù)獲取，計算這K個組合的平均識別率）
def totalAlgorithm_1():
 #獲取各個分類器
 clf_KNN = KNN()
 clf_LDA = LDA()
 clf_SVM = SVM()
 clf_LR = LR()
 clf_RF = RF()
 clf_NBC = native_bayes_classifier()
 clf_DTC = decision_tree_classifier()
 clf_GBDT = gradient_boosting_classifier()
 #獲取訓練集和測試集
 setDict = getData_3()
 setNums = len(setDict.keys()) / 4 #一共生成了setNums個訓練集和setNums個測試集，它們之間是一一對應關系
 #定義變量，用于將每個分類器的所有識別率累加
 KNN_rate = 0.0
 LDA_rate = 0.0
 SVM_rate = 0.0
 LR_rate = 0.0
 RF_rate = 0.0
 NBC_rate = 0.0
 DTC_rate = 0.0
 GBDT_rate = 0.0
 for i in range(1, int(setNums + 1)):
  trainMatrix = setDict[str(i) + 'train']
  trainClass = setDict[str(i) + 'trainclass']
  testMatrix = setDict[str(i) + 'test']
  testClass = setDict[str(i) + 'testclass']
  #輸入訓練樣本
  clf_KNN.fit(trainMatrix, trainClass)
  clf_LDA.fit(trainMatrix, trainClass)
  clf_SVM.fit(trainMatrix, trainClass)
  clf_LR.fit(trainMatrix, trainClass)
  clf_RF.fit(trainMatrix, trainClass)
  clf_NBC.fit(trainMatrix, trainClass)
  clf_DTC.fit(trainMatrix, trainClass)
  clf_GBDT.fit(trainMatrix, trainClass)
  #計算識別率
  KNN_rate += getRecognitionRate(clf_KNN.predict(testMatrix), testClass)
  LDA_rate += getRecognitionRate(clf_LDA.predict(testMatrix), testClass)
  SVM_rate += getRecognitionRate(clf_SVM.predict(testMatrix), testClass)
  LR_rate += getRecognitionRate(clf_LR.predict(testMatrix), testClass)
  RF_rate += getRecognitionRate(clf_RF.predict(testMatrix), testClass)
  NBC_rate += getRecognitionRate(clf_NBC.predict(testMatrix), testClass)
  DTC_rate += getRecognitionRate(clf_DTC.predict(testMatrix), testClass)
  GBDT_rate += getRecognitionRate(clf_GBDT.predict(testMatrix), testClass)
 #輸出各個分類器的平均識別率（K個訓練集測試集，計算平均）
 print
 print
 print
 print('K Nearest Neighbor mean recognition rate: ', KNN_rate / float(setNums))
 print('Linear Discriminant Analysis mean recognition rate: ', LDA_rate / float(setNums))
 print('Support Vector Machine mean recognition rate: ', SVM_rate / float(setNums))
 print('Logistic Regression mean recognition rate: ', LR_rate / float(setNums))
 print('Random Forest mean recognition rate: ', RF_rate / float(setNums))
 print('Native Bayes Classifier mean recognition rate: ', NBC_rate / float(setNums))
 print('Decision Tree Classifier mean recognition rate: ', DTC_rate / float(setNums))
 print('Gradient Boosting Decision Tree mean recognition rate: ', GBDT_rate / float(setNums))

#“主”函數(shù)2（每類前x%作為訓練集，剩余作為測試集，即數(shù)據(jù)集用getData_2()方法獲取，計算識別率）
def totalAlgorithm_2():
 #獲取各個分類器
 clf_KNN = KNN()
 clf_LDA = LDA()
 clf_SVM = SVM()
 clf_LR = LR()
 clf_RF = RF()
 clf_NBC = native_bayes_classifier()
 clf_DTC = decision_tree_classifier()
 clf_GBDT = gradient_boosting_classifier()
 #獲取訓練集和測試集
 T = getData_2()
 trainMatrix, trainClass, testMatrix, testClass = T[0], T[1], T[2], T[3]
 #輸入訓練樣本
 clf_KNN.fit(trainMatrix, trainClass)
 clf_LDA.fit(trainMatrix, trainClass)
 clf_SVM.fit(trainMatrix, trainClass)
 clf_LR.fit(trainMatrix, trainClass)
 clf_RF.fit(trainMatrix, trainClass)
 clf_NBC.fit(trainMatrix, trainClass)
 clf_DTC.fit(trainMatrix, trainClass)
 clf_GBDT.fit(trainMatrix, trainClass)
 #輸出各個分類器的識別率
 print('K Nearest Neighbor recognition rate: ', getRecognitionRate(clf_KNN.predict(testMatrix), testClass))
 print('Linear Discriminant Analysis recognition rate: ', getRecognitionRate(clf_LDA.predict(testMatrix), testClass))
 print('Support Vector Machine recognition rate: ', getRecognitionRate(clf_SVM.predict(testMatrix), testClass))
 print('Logistic Regression recognition rate: ', getRecognitionRate(clf_LR.predict(testMatrix), testClass))
 print('Random Forest recognition rate: ', getRecognitionRate(clf_RF.predict(testMatrix), testClass))
 print('Native Bayes Classifier recognition rate: ', getRecognitionRate(clf_NBC.predict(testMatrix), testClass))
 print('Decision Tree Classifier recognition rate: ', getRecognitionRate(clf_DTC.predict(testMatrix), testClass))
 print('Gradient Boosting Decision Tree recognition rate: ', getRecognitionRate(clf_GBDT.predict(testMatrix), testClass))

if __name__ == '__main__':
 print('K個訓練集和測試集的平均識別率')
 totalAlgorithm_1()
 print('每類前x%訓練，剩余測試，各個模型的識別率')
 totalAlgorithm_2()
 selectRFParam()
 print('隨機森林參數(shù)調優(yōu)完成！')