Python sklearn包的使用示例以及参数调优示例

coding=utf-8

!/usr/bin/env python

””’
【说明】
1.当前sklearn版本0.18
2.sklearn自带的鸢尾花数据集样例：
（1）样本特征矩阵（类型：numpy.ndarray）
[[ 6.7 3. 5.2 2.3]
[ 6.3 2.5 5. 1.9]
[ 6.5 3. 5.2 2. ]
[ 6.2 3.4 5.4 2.3]
[ 5.9 3. 5.1 1.8]]
每行是一个样本，矩阵行数=样本总数，矩阵列数=每个样本特征数
（2）样本类别矩阵（类型：numpy.ndarray）
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2]
每个元素对应一个样本的类标
3.本地excel表的数据集样例：
class0 p1 p2 p3 p4 p5 p6 p7
0 0 0 0 1 0 0 0
0 5 9 10 10 0 1 1
0 0 1 1 0 0 1 0
0 0 1 1 0 0 1 0
每行是一个样本，每行第一个元素是样本所属类别，后续元素是样本的特征
”’
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

读取sklearn自带的数据集（鸢尾花）

def getData_1():
iris = datasets.load_iris()
X = iris.data #样本特征矩阵，150*4矩阵，每行一个样本，每个样本维度是4
y = iris.target #样本类别矩阵，150维行向量，每个元素代表一个样本的类别

读取本地excel表格内的数据集（抽取每类60%样本组成训练集，剩余样本组成测试集）

返回一个元祖，其内有4个元素（类型均为numpy.ndarray）：

（1）归一化后的训练集矩阵，每行为一个训练样本，矩阵行数=训练样本总数，矩阵列数=每个训练样本的特征数

（2）每个训练样本的类标

（3）归一化后的测试集矩阵，每行为一个测试样本，矩阵行数=测试样本总数，矩阵列数=每个测试样本的特征数

（4）每个测试样本的类标

【注】归一化采用“最大最小值”方法。

def getData_2():
fPath = ‘F:/cleanData_dropSJS.csv’
if os.path.exists(fPath):
data = pd.read_csv(fPath,header=None,skiprows=1,names=[‘class0’,’pixel0’,’pixel1’,’pixel2’,’pixel3’,’pixel4’,’pixel5’, ‘pixel6’])
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() #归一化
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!’

读取本地excel表格内的数据集（每类随机生成K个训练集和测试集的组合）

【K的含义】假设一共有1000个样本，K取10，那么就将这1000个样本切分10份（一份100个），那么就产生了10个测试集

对于每一份的测试集，剩余900个样本即作为训练集

结果返回一个字典：键为集合编号（1train, 1trainclass, 1test, 1testclass, 2train, 2trainclass, 2test, 2testclass…），值为数据

其中1train和1test为随机生成的第一组训练集和测试集（1trainclass和1testclass为训练样本类别和测试样本类别），其他以此类推

def getData_3():
fPath = ‘F:/cleanData_dropSJS.csv’
if os.path.exists(fPath):
#读取csv文件内的数据，
dataMatrix = np.array(pd.read_csv(fPath,header=None,skiprows=1,names=[‘class0’,’pixel0’,’pixel1’,’pixel2’,’pixel3’,’pixel4’,’pixel5’, ‘pixel6’]))
#获取每个样本的特征以及类标
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) #二维矩阵，一行是一个样本，行数=样本总数，列数=样本特征数
classM = np.array(sampleClass) #一维列向量，每个元素对应每个样本所属类别
#调用StratifiedKFold方法生成训练集和测试集
skf = StratifiedKFold(n_splits = 10)
setDict = {} #创建字典，用于存储生成的训练集和测试集
count = 1
for trainI, testI in skf.split(sampleM, classM):
trainSTemp = [] #用于存储当前循环抽取出的训练样本数据
trainCTemp = [] #用于存储当前循环抽取出的训练样本类标
testSTemp = [] #用于存储当前循环抽取出的测试样本数据
testCTemp = [] #用于存储当前循环抽取出的测试样本类标
#生成训练集
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)
#生成测试集
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近邻（K Nearest Neighbor）

def KNN():
clf = neighbors.KNeighborsClassifier()
return clf

线性鉴别分析（Linear Discriminant Analysis）

def LDA():
clf = LinearDiscriminantAnalysis()
return clf

支持向量机（Support Vector Machine）

def SVM():
clf = svm.SVC()
return clf

逻辑回归（Logistic Regression）

def LR():
clf = LogisticRegression()
return clf

随机森林决策树（Random Forest）

def RF():
clf = RandomForestClassifier()
return clf

多项式朴素贝叶斯分类器

def native_bayes_classifier():
clf = MultinomialNB(alpha = 0.01)
return clf

决策树

def decision_tree_classifier():
clf = tree.DecisionTreeClassifier()
return clf

GBDT

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

计算识别率

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函数，将调参的详细结果存储到本地F盘（路径可自行修改，其中n_top是指定输出前多少个最优参数组合以及该组合的模型得分）

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()

自动调参（以随机森林为例）

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() #获取数据集
grid_search.fit(T[0], T[1]) #传入训练集矩阵和训练样本类标
print(“GridSearchCV took %.2f seconds for %d candidate parameter settings.”
% (time() - start, len(grid_search.cv_results_[‘params’])))
report(grid_search.cv_results_)

“主”函数1（KFold方法生成K个训练集和测试集，即数据集采用getData_3()函数获取，计算这K个组合的平均识别率）

def totalAlgorithm_1():
#获取各个分类器
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()
#获取训练集和测试集
setDict = getData_3()
setNums = len(setDict.keys()) / 4 #一共生成了setNums个训练集和setNums个测试集，它们之间是一一对应关系
#定义变量，用于将每个分类器的所有识别率累加
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, setNums + 1):
trainMatrix = setDict[str(i) + ‘train’]
trainClass = setDict[str(i) + ‘trainclass’]
testMatrix = setDict[str(i) + ‘test’]
testClass = setDict[str(i) + ‘testclass’]
#输入训练样本
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)
#计算识别率
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)
#输出各个分类器的平均识别率（K个训练集测试集，计算平均）
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”函数2（每类前x%作为训练集，剩余作为测试集，即数据集用getData_2()方法获取，计算识别率）

def totalAlgorithm_2():
#获取各个分类器
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()
#获取训练集和测试集
T = getData_2()
trainMatrix, trainClass, testMatrix, testClass = T[0], T[1], T[2], T[3]
#输入训练样本
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)
#输出各个分类器的识别率
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个训练集和测试集的平均识别率’)
totalAlgorithm_1()
print(‘每类前x%训练，剩余测试，各个模型的识别率’)
totalAlgorithm_2()
selectRFParam()
print(‘随机森林参数调优完成！’)

”’
【输出结果】
K个训练集和测试集的平均识别率
(‘K Nearest Neighbor mean recognition rate: ‘, 0.48914314291650945)
(‘Linear Discriminant Analysis mean recognition rate: ‘, 0.5284076063968655)
(‘Support Vector Machine mean recognition rate: ‘, 0.5271199740575014)
(‘Logistic Regression mean recognition rate: ‘, 0.5620828985391165)
(‘Random Forest mean recognition rate: ‘, 0.512993404168108)
(‘Native Bayes Classifier mean recognition rate: ‘, 0.4467074333715003)
(‘Decision Tree Classifier mean recognition rate: ‘, 0.47351209424438706)
(‘Gradient Boosting Decision Tree mean recognition rate: ‘, 0.5603633086892212)
每类前x%训练，剩余测试，各个模型的识别率
(‘K Nearest Neighbor recognition rate: ‘, 0.9892818863879957)
(‘Linear Discriminant Analysis recognition rate: ‘, 1.0)
(‘Support Vector Machine recognition rate: ‘, 0.8928188638799571)
(‘Logistic Regression recognition rate: ‘, 0.8494105037513398)
(‘Random Forest recognition rate: ‘, 0.9801714898177921)
(‘Native Bayes Classifier recognition rate: ‘, 0.7604501607717041)
(‘Decision Tree Classifier recognition rate: ‘, 1.0)
(‘Gradient Boosting Decision Tree recognition rate: ‘, 1.0)
GridSearchCV took 69.51 seconds for 288 candidate parameter settings.
随机森林参数调优完成！