k-近邻分类的Python实现

参见《机器学习实战》

 # -*- coding:cp936 -*-
 #===============================================================================
 # 设计KNN最近邻分类器：
 #     找出每个元素在数据集中的最近邻的K个数据，统计这K个数据所属的类，所属类最多的那个类就是该元素所属的类
 #===============================================================================
 import numpy as np

 def loadHaiLunData(f_name):
     with open(f_name) as fHandle:
         fLines = fHandle.readlines()
         dataLines = len(fLines)
         label  = []
         dataSetMat = np.zeros((dataLines,3))
         for i in range(dataLines):
             lineList = fLines[i].strip().split('\t')
             dataSetMat[i,:] = lineList[0:3]
             label.append(int(lineList[-1]))
         return dataSetMat,label

 def dataNorm(dataSet):
     numOfEle = dataSet.shape[0]
     minEle = dataSet.min(0)
     maxEle = dataSet.max(0)
     normedData = (dataSet-np.tile(minEle,(numOfEle,1)))/np.tile(maxEle-minEle,(numOfEle,1))
     return normedData

 def classifyKnn(inX, dataSet, label, k):
     #===========================================================================
     # inX：输入向量
     # dataSet:保存数据特征的数组，每一行为若干个特征的参数，与label对应
     # label：表明当前这个数据集中的每一个元素属于哪一类
     # k：设定最近邻的个数
     #===========================================================================

     #首先对数据集进行归一化
 #     dataSet = dataNorm(dataSet)
     numOfEle = dataSet.shape[0]
     index = 0
     diffDistance = dataSet - np.tile(inX, (numOfEle,1))
     diffDistance = diffDistance**2
     squareDistance = diffDistance.sum(1)
 #     squareDistance = squareDistance**0.5
     knnIndex = squareDistance.argsort()
     #统计最近的k个近邻的label，看哪个label类别最多就可将该训练元素判为对应类
     staticDict = {}
     for i in range(k):
         staticDict[label[knnIndex[i]]]=staticDict.get(label[knnIndex[i]],0)+1
     itemList = staticDict.items()
     argmax = np.argmax(itemList, axis = 0)
     return itemList[argmax[1]][0]

 def testHaiLunClassify(k = 3, hRatio = 0.5):
     dataSet,label = loadHaiLunData('datingTestSet2.txt')
 #     hRatio = 0.5
     totalNum = dataSet.shape[0]
     testNum = int(totalNum*hRatio)
     dataNormed = dataNorm(dataSet)
     errorClass = 0
     for i in range(testNum):
         classRes = classifyKnn(dataNormed[i,:], dataNormed[testNum:,:], label[testNum:], k)
         if classRes != label[i]:
             errorClass += 1
 #             print "classify error, No. %d should be label %d but got %d"%(i, label[i],classRes)
     errorRate = errorClass/float(testNum)
 #     print
 #     print "Error rate: %f"%(errorRate)
     return errorRate

 if __name__ == '__main__':
     errorList = []
     kRange = range(1,50,1)
     for k in kRange:
         errorList.append(testHaiLunClassify(k))
     print errorList
     import matplotlib.pyplot as  plt
     fig = plt.figure(1)
 #     ax  = fig.add_subplot(111)
     plt.plot(kRange, errorList,'rs-')
     plt.show()