k-Nearest Neighbors 实战2 使用kNN算法改进约会网站的配对结果

本文《machine learning in action》学习笔记
数据源码可以在这里获取 ：https://www.manning.com/books/machine-learning-in-action

---

这里Python 3+的code

from numpy import *
import matplotlib.pyplot as plt
import operator

def kNNClassify(inX, dataSet, labels, k):
    '''put the kNN classification algorithm into action'''

    dataSetSize = dataSet.shape[0]
    diffMax = tile(inX,(dataSetSize,1)) - dataSet
    sqDiffMax = diffMax ** 2
    sqDistances = sqDiffMax.sum(axis=1)
    distances = sqDistances**0.5
    # argsort 返回由大到小的索引值
    sortedDistIndicies = distances.argsort()
    classCount= {}

    for i in range(k):
        # 找到最大索引值对应数据的label
        voteIlabel = labels[sortedDistIndicies[i]]
         # returns a value for the given key
        classCount[voteIlabel] = classCount.get(voteIlabel,0) + 1
    # 按照键值的大小排列
    sortedClassCount = sorted(classCount.items(), key = operator.itemgetter(1), reverse = True)
    return sortedClassCount[0][0]



def file2matrix(filename):
    """process the text information"""
    fr = open(filename)
    arrayofLines = fr.readlines()
    fr.close()
    numberofLines = len(arrayofLines)
    returnMat = zeros((numberofLines, 3))
    classLabelVector = []

    index = 0
    for line in arrayofLines:
        # Python strip() 方法用于移除字符串头尾指定的字符（默认为空格或换行符）
        line = line.strip()
        # split()通过指定分隔符对字符串进行切片，这里使用tab
        listFromLine = line.split('	')
        returnMat[index,:] = listFromLine[0:3]
        classLabelVector.append(int(listFromLine[-1]))
        index += 1
    return returnMat, classLabelVector


def autonorm(dataset):
    """归一化"""
    # 求每一列最小值和最大值
    minvalue = dataset.min(0)
    maxvalue = dataset.max(0)
    ranges = maxvalue - minvalue
    # 使用shape获取dataset的shape
    normdataset = zeros(shape(dataset))
    # shape[0] 获取第一行元素个数
    m = dataset.shape[0]
    # 使用tile函数将变量内容复制成输入矩阵同样大小的矩阵做矩阵的减法
    normdataset = dataset - tile(minvalue, (m,1))
    # 做矩阵的除法
    normdataset = normdataset / tile(ranges, (m,1))
    return normdataset, ranges, minvalue

def datingclasstest():
    horatio = 0.1
    datingdatamat, datinglabel = file2matrix("datingTestset2.txt")
    normat, ranges, minvalues = autonorm(datingdatamat)
    m = normat.shape[0]
    numtestvec = int(m*horatio)
    errorcount = 0.0
    for i in range(numtestvec):
        #classifierresult = classify0(normat[i,:], normat[numtestvec:m,:], datinglabel[numtestvec:m,:], 3)
        classifierresult = kNNClassify(normat[i, :], normat[numtestvec:m], datinglabel[numtestvec:m], 3)
        print("No.%d test data, the classifier came back with : %d, the real answeris: %d" %(i, classifierresult, datinglabel[i]))
        if (classifierresult != datinglabel[i]):
            errorcount += 1.0
    print ("the total error rate is: %f" % (errorcount/float(numtestvec)))



if __name__ == '__main__':
    datingdatamat, datinglabel = file2matrix('datingTestSet2.txt')
    normdataset, ranges, minvalue = autonorm(datingdatamat)
    print(normdataset)
    print("ranges = ", ranges)
    print("minvalue = ", minvalue)

    fig = plt.figure()
    ax = fig.add_subplot(111)
    # 画散点图 scatter
    ax.scatter(datingdatamat[:,1], datingdatamat[:,2], 15.0*array(datinglabel), 15.0*array(datinglabel))
    plt.show()

    datingclasstest()

读取TXT数据

def file2matrix(filename):
    """process the text information"""
    fr = open(filename)
    arrayofLines = fr.readlines()
    numberofLines = len(arrayofLines)
    returnMat = zeros((numberofLines, 3))
    classLabelVector = []

    index = 0
    for line in arrayofLines:
        # Python strip() 方法用于移除字符串头尾指定的字符（默认为空格或换行符）
        line = line.strip()
        # split()通过指定分隔符对字符串进行切片，这里使用tab
        listFromLine = line.split('	')
        returnMat[index:1] = listFromLine[0:3]
        classLabelVector.append(int(listFromLine[-1]))
        index += 1
    return returnMat, classLabelVector

normalization result：

plot：

test result：

结果与书中的结果并不一致，kNN这么简单的算法，其结果应该一致才对。为什么？