机器学习实战--logistic回归

  1 #encoding:utf-8
  2 from numpy import *
  3 
  4 def loadDataSet():  #加载数据
  5     dataMat = [];
  6     labelMat = []
  7     fr = open(‘testSet.txt‘)
  8     for line in fr.readlines():
  9         lineArr = line.strip().split()
 10         dataMat.append([1.0, float(lineArr[0]), float(lineArr[1])])
 11         labelMat.append(int(lineArr[2]))
 12     return dataMat, labelMat
 13 
 14 
 15 def sigmoid(inX):    #得到sigmoid函数值
 16     return 1.0 / (1 + exp(-inX))
 17 
 18 
 19 def gradAscent(dataMatIn, classLabels):
 20     dataMatrix = mat(dataMatIn)  # 转化为numpy矩阵
 21     labelMat = mat(classLabels).transpose()  # 转化为numpy矩阵，并转置
 22     m, n = shape(dataMatrix)
 23     alpha = 0.001
 24     maxCycles = 500
 25     weights = ones((n, 1))
 26     for k in range(maxCycles):  # 迭代maxCycles次   梯度上升算法
 27         h = sigmoid(dataMatrix * weights)
 28         error = (labelMat - h)
 29         weights = weights + alpha * dataMatrix.transpose() * error  # 为什么这么做？参考附件，或者http://download.csdn.net/detail/lewsn2008/6547463，总结的非常好
 30     return weights
 31 
 32 
 33 def plotBestFit(weights):   #画出数据集和最佳拟合曲线
 34     import matplotlib.pyplot as plt
 35     dataMat, labelMat = loadDataSet()
 36     dataArr = array(dataMat)
 37     #weights = weights.getA()
 38     n = shape(dataArr)[0]
 39     xcord1 = [];
 40     ycord1 = []
 41     xcord2 = [];
 42     ycord2 = []
 43     for i in range(n):
 44         if int(labelMat[i]) == 1:
 45             xcord1.append(dataArr[i, 1]);
 46             ycord1.append(dataArr[i, 2])
 47         else:
 48             xcord2.append(dataArr[i, 1]);
 49             ycord2.append(dataArr[i, 2])
 50     fig = plt.figure()
 51     ax = fig.add_subplot(111)
 52     ax.scatter(xcord1, ycord1, s=30, c=‘red‘, marker=‘s‘)
 53     ax.scatter(xcord2, ycord2, s=30, c=‘green‘)
 54     x = arange(-3.0, 3.0, 0.1)
 55     y = (-weights[0] - weights[1] * x) / weights[2]
 56     ax.plot(x, y)
 57     plt.xlabel(‘X1‘);
 58     plt.ylabel(‘X2‘);
 59     plt.show()
 60 
 61 # 梯度上升算法在每次更新回归系数时都需要遍历整个数据集，
 62 # 该方法在处理100个左右的数据集尚可，但如果数据量增大，那该方法的计算量就太大了，
 63 # 有一种改进方法是一次仅用一个样本点来更新回归系数，该方法称为随机梯度上升算法，
 64 # 由于可以在新样本到来时对分类器进行增量式更新，因而随机梯度上升算法是一个在线学习算法。
 65 def stocGradAscent0(dataMatrix, classLabels):   #随机梯度上升算法
 66     m, n = shape(dataMatrix)
 67     alpha = 0.01
 68     weights = ones(n)  # initialize to all ones
 69     for i in range(m):
 70         h = sigmoid(sum(dataMatrix[i] * weights))
 71         error = classLabels[i] - h
 72         weights = weights + alpha * error * dataMatrix[i]
 73     return weights
 74 
 75 def useStocGradAscent0():    #测试随机梯度上升算法
 76     dataMat, labelMat = loadDataSet()
 77     weights = stocGradAscent0(array(dataMat), labelMat)
 78     plotBestFit(weights)
 79 
 80 def useStocGradAscent1():    #测试改进的随机梯度上升算法
 81     dataMat, labelMat = loadDataSet()
 82     weights = stocGradAscent1(array(dataMat), labelMat)
 83     plotBestFit(weights)
 84 
 85 def stocGradAscent1(dataMatrix, classLabels, numIter=150):   #改进的随机梯度上升算法
 86     m, n = shape(dataMatrix)
 87     weights = ones(n)
 88     for j in range(numIter):
 89         dataIndex = range(m)
 90         for i in range(m):
 91             alpha = 4 / (1.0 + j + i) + 0.0001  # alpha每次调整
 92             randIndex = int(random.uniform(0, len(dataIndex)))  # 随机选取更新
 93             h = sigmoid(sum(dataMatrix[randIndex] * weights))
 94             error = classLabels[randIndex] - h
 95             weights = weights + alpha * error * dataMatrix[randIndex]
 96             del (dataIndex[randIndex])
 97     return weights
 98 
 99 
100 def classifyVector(inX, weights):   #得到类别
101     prob = sigmoid(sum(inX * weights))
102     if prob > 0.5:
103         return 1.0
104     else:
105         return 0.0
106 
107 
108 def colicTest():
109     frTrain = open(‘horseColicTraining.txt‘);   #读取文件
110     frTest = open(‘horseColicTest.txt‘)
111     trainingSet = [];
112     trainingLabels = []
113     for line in frTrain.readlines():
114         currLine = line.strip().split(‘\t‘)
115         lineArr = []
116         for i in range(21):
117             lineArr.append(float(currLine[i]))
118         trainingSet.append(lineArr)
119         trainingLabels.append(float(currLine[21]))
120     trainWeights = stocGradAscent1(array(trainingSet), trainingLabels, 1000)  #用改进的随机梯度上升法训练
121     errorCount = 0;
122     numTestVec = 0.0
123     for line in frTest.readlines():
124         numTestVec += 1.0
125         currLine = line.strip().split(‘\t‘)
126         lineArr = []
127         for i in range(21):
128             lineArr.append(float(currLine[i]))
129         if int(classifyVector(array(lineArr), trainWeights)) != int(currLine[21]):    #对测试集分类，并判断是否正确
130             errorCount += 1
131     errorRate = (float(errorCount) / numTestVec)
132     print "the error rate of this test is: %f" % errorRate
133     return errorRate
134 
135 
136 def multiTest():   #测试
137     numTests = 10;
138     errorSum = 0.0
139     for k in range(numTests):
140         errorSum += colicTest()
141     print "after %d iterations the average error rate is: %f" % (numTests, errorSum / float(numTests))
142 
143 if __name__ == ‘__main__‘:
144     # dataMat, labelMat = loadDataSet()
145     # plotBestFit(gradAscent(dataMat, labelMat).getA())
146     #useStocGradAscent1()
147     multiTest()