05 逻辑回归

from numpy import *


def loadDataSet():   #加载数据
    dataMat = []; labelMat = []
    fr = open('testSet.txt')    #打开指定文件
    for line in fr.readlines():
        lineArr = line.strip().split()
        dataMat.append([1.0, float(lineArr[0]), float(lineArr[1])])   #将数据放入列表中
        labelMat.append(int(lineArr[2]))           #将标签加入到数据集合中
    return dataMat,labelMat    #返回列表内容


def sigmoid(inX):         #sigmoid函数，阶跃函数
    return 1.0/(1+exp(-inX))


def gradAscent(dataMatIn, classLabels):     #梯度上升算法
    dataMatrix = mat(dataMatIn)             #转化为numpy矩阵
    labelMat = mat(classLabels).transpose() #转化为numpy矩阵
    m,n = shape(dataMatrix)      #shape函数求得矩阵的行和列，m表示行，n表示列
    alpha = 0.001          #梯度
    maxCycles = 500        #迭代的次数
    weights = ones((n,1))   #创建一个n行1列的矩阵
    for k in range(maxCycles):              #heavy on matrix operations
        h = sigmoid(dataMatrix*weights)     #放入到sigmoid函数中计算
        error = (labelMat - h)              #计算预测值与真实值之间的距离
        weights = weights + alpha * dataMatrix.transpose()* error   #transpose表示转置
    return weights   #返回拟合的参数系数


def plotBestFit(weights):          #一个绘图的函数，具体看效果
    import matplotlib.pyplot as plt
    dataMat,labelMat=loadDataSet()
    dataArr = array(dataMat)
    n = shape(dataArr)[0] 
    xcord1 = []; ycord1 = []
    xcord2 = []; ycord2 = []
    for i in range(n):
        if int(labelMat[i])== 1:
            xcord1.append(dataArr[i,1]); ycord1.append(dataArr[i,2])
        else:
            xcord2.append(dataArr[i,1]); ycord2.append(dataArr[i,2])
    fig = plt.figure()
    ax = fig.add_subplot(111)
    ax.scatter(xcord1, ycord1, s=30, c='red', marker='s')
    ax.scatter(xcord2, ycord2, s=30, c='green')
    x = arange(-3.0, 3.0, 0.1)
    y = (-weights[0]-weights[1]*x)/weights[2]
    ax.plot(x, y)
    plt.xlabel('X1'); plt.ylabel('X2');
    plt.show()


def stocGradAscent0(dataMatrix, classLabels): #随机梯度上升1.0版本，每次更新样本参数时，只选择一个样本进行更新
    m,n = shape(dataMatrix)
    alpha = 0.01
    weights = ones(n)   #initialize to all ones,创建一个n行1列的矩阵，初始化为1
    for i in range(m):
        h = sigmoid(sum(dataMatrix[i]*weights))
        error = classLabels[i] - h
        weights = weights + alpha * error * dataMatrix[i] #每次只选择一个样本进行更新
    return weights  #返回最优参数系数


def stocGradAscent1(dataMatrix, classLabels, numIter=150):  #随机梯度上升2.0版本
    m,n = shape(dataMatrix)
    weights = ones(n)   #initialize to all ones
    for j in range(numIter):
        dataIndex =list(range(m))
        for i in range(m):
            alpha = 4/(1.0+j+i)+0.0001    #apha decreases with iteration, does not ，动态变化的步长
            randIndex = int(random.uniform(0,len(dataIndex)))#go to 0 because of the constant，随机选择一个样本用来更新weights
            h = sigmoid(sum(dataMatrix[randIndex]*weights))
            error = classLabels[randIndex] - h
            weights = weights + alpha * error * dataMatrix[randIndex]
            del(dataIndex[randIndex])   #删除已经用来更新weight之后的样本
    return weights


def classifyVector(inX, weights):
    prob = sigmoid(sum(inX*weights))   #sigmoid函数
    if prob > 0.5: return 1.0       #大于0.5，分类为1
    else: return 0.0               #小于0.5，分类为0


def colicTest():      #测试的函数
    frTrain = open('horseColicTraining.txt'); frTest = open('horseColicTest.txt')
    trainingSet = []; trainingLabels = []
    for line in frTrain.readlines():
        currLine = line.strip().split('\t')
        lineArr =[]
        for i in range(21):
            lineArr.append(float(currLine[i]))
        trainingSet.append(lineArr)
        trainingLabels.append(float(currLine[21]))
    trainWeights = stocGradAscent1(array(trainingSet), trainingLabels, 1000)
    errorCount = 0; numTestVec = 0.0
    for line in frTest.readlines():
        numTestVec += 1.0
        currLine = line.strip().split('\t')
        lineArr =[]
        for i in range(21):
            lineArr.append(float(currLine[i]))
        if int(classifyVector(array(lineArr), trainWeights))!= int(currLine[21]):
            errorCount += 1
    errorRate = (float(errorCount)/numTestVec)
    print ("the error rate of this test is: %f" % errorRate)
    return errorRate


def multiTest():
    numTests = 10; errorSum=0.0
    for k in range(numTests):
        errorSum += colicTest()
    print ("after %d iterations the average error rate is: %f" % (numTests, errorSum/float(numTests)))




#main函数


if __name__ == "__main__":
    dataMat, labelMat=loadDataSet()
    # weights=stocGradAscent1(array(dataMat),labelMat,500)
    # plotBestFit(weights)
    multiTest()


这是一个逻辑回归的代码，使用的是梯度上升的算法，最终的是sigmoid函数，记住这个函数，它有着单位阶跃的性质。