KNN算法是机器学习经典十大算法之一，简单易懂。这里给出KNN的实现，由两个版本：
1.机器学习实战上作者的实现版本，我自己又敲了一遍感觉还是蛮有收获的；
2.用自己的理解的一个实现，主要的区别就是效率没有第一个高，因为第一个大量使用矩阵向量的运算，速度比较快，还有就是作者的代码比较简介好看。自己的代码就是比较好懂。
1.《机器学习实战》的代码
1 # -*- coding: utf-8 -*-
2
3 '''
4 function: 根据《实战》实现KNN算法,用于约会网站进行匹配
5 date: 2017.8.5
6 '''
7
8 from numpy import *
9 import matplotlib.pyplot as plt
10 import operator
11
12 #产生数据集和标签
13 def createDataSet():
14 group = array([[1.0,1.1], [1.0,1.0], [0,0], [0,0.1]])
15 labels = ['A', 'A', 'B', 'B']
16 return group, labels
17
18 #knn分类算法
19 def classify0(inX, dataSet, labels, k):
20 dataSetSize = dataSet.shape[0] #返回m,n,shape[0] == m 行数,代表由几个训练数据
21
22 #计算距离;将测试向量 inX 在咧方向上重复一次，行方向重复m次，与对应的训练数据相减
23 diffMat = tile(inX, (dataSetSize, 1)) - dataSet
24
25 sqDiffMat = diffMat**2 #平方的意思，不能用^
26 sqDistances = sqDiffMat.sum(axis=1) #axis=1表示按行相加，axis=0表示案列相加
27 distances = sqDistances**0.5 #测试点距离所有训练点的距离向量集合
28
29 sortedDistIndices = distances.argsort() #距离从小到大排序，返回的是排序后的下表组成的数组；
30 classCount = {}
31
32 for i in range(k):
33 voteIlabel = labels[sortedDistIndices[i]]
34 classCount[voteIlabel] = classCount.get(voteIlabel,0) + 1 #get函数返回value，若不存在则返回0
35
36 #items()返回key,value,itemgetter(1)，按照第二个元素进行排序，从大到小
37 sortedClassCount = sorted(classCount.items(), key=operator.itemgetter(1),reverse=True)
38
39 return sortedClassCount[0][0]
40
41 #将文本记录转换为nmpy的解析程序
42 def file2matrix(filename):
43 fr = open(filename)
44 arrayOLines = fr.readlines()
45 numberOfLines = len(arrayOLines)
46 returnMat = zeros((numberOfLines,3)) #返回的矩阵 是： m*3,3个特征值
47 classLabelVector = []
48 index = 0
49 for line in arrayOLines:
50 line = line.strip() #去掉换行符
51 listFromLine = line.split(' ') #将每一行用tab分割
52 returnMat[index,:] = listFromLine[0:3] #每行前三个是特征
53 classLabelVector.append(int(listFromLine[-1])) #最后一列是标签，并且告诉list是int类型
54 index += 1
55 return returnMat, classLabelVector
56
57 #draw the point
58 def drawPoint(datingDatamat, datingLabels):
59 fig = plt.figure()
60 ax = fig.add_subplot(111)
61
62 #总共三个特征，可以选择其中任意两个来画图，看哪个区分效果好
63 ax.scatter(datingDatamat[:,0],datingDatamat[:,1],15.0*array(datingLabels),15.0*array(datingLabels))
64 #ax.scatter(datingDatamat[:,1],datingDatamat[:,2],15.0*array(datingLabels),15.0*array(datingLabels))
65 plt.show()
66
67
68 #对数据集归一化处理 autoNormanization
69 def autoNorm(dataSet):
70 minVals = dataSet.min(0) #get the minimal row
71 maxVals = dataSet.max(0) #get the maximal row
72 ranges = maxVals - minVals + ones_like(maxVals) #个人感觉这里可能由除〇的风险，加一效果可能好一点
73 normDataSet = zeros(shape(dataSet)) #the same size of original dataset: m*n
74 m = dataSet.shape[0]
75 normDataSet = dataSet - tile(minVals, (m,1)) #repeat the minimal row m times on the row
76 normDataSet = dataSet / tile(ranges, (m,1))
77 return normDataSet, ranges, minVals
78
79 #分类器针对约会网站的测试代码
80 def datingClassTest():
81 haRatio = 0.10
82 datingDataMat, datingLabels = file2matrix('datingTestSet2.txt')
83 normMat, ranges, minVals = autoNorm(datingDataMat)
84 m = normMat.shape[0] #获得矩阵的行数，数据集的个数
85 numTestVecs = int(m*haRatio) #用十分之一的数据来做测试，剩下的作为训练数据
86 errorCount = 0.0
87 for i in range(numTestVecs): #对每个测试用例开始计算
88 #0~numtestvecs都是用来测试的，每次测试一行，从numtestvecs~m都是训练数据
89 classifierResult = classify0(normMat[i,:],normMat[numTestVecs:m,:],datingLabels[numTestVecs:m],3)
90 print('the classifier came back with : %d, the real answer is : %d'
91 % (classifierResult,datingLabels[i]))
92 if(classifierResult != datingLabels[i]):
93 errorCount += 1
94 print('the total error rate is: %f' % (errorCount / float(numTestVecs)))
95
96
97 #让用户输入自己的一些信息，用输入的信息进行分类
98 def classifyPerson():
99 resultList = ['not at all', 'in small doses', 'in large doses']
100 percentTats = float(input('persentage of time spent playing video games?'))
101 ffMiles = float(input('frequent fliers miles each year?'))
102 iceCream = float(input('liters of ice cream consumed per year?'))
103 datingDatamat, datingLabels = file2matrix('datingTestSet2.txt')
104 normMat, ranges, minVals = autoNorm(datingDatamat)
105 inArr = array([percentTats, ffMiles, iceCream])
106 #对输入的一个测试数据进行归一化然后分类
107 classifierResult = classify0((inArr - minVals) / ranges, normMat, datingLabels, 3)
108 print('you will probably like this person: ', resultList[classifierResult - 1])
109 print(classifierResult)
110
111 # #测试分类效果和加载数据的程序
112 # group, labels = createDataSet()
113 # result = classify0([0,0], group, labels, 3)
114 # print(result)
115
116 # #get numpy metrix from file
117 # returnMat, classLabelVector = file2matrix('datingTestSet2.txt')
118 # print(returnMat)
119 # print(classLabelVector)
120
121 # #auto normanization
122 # normDataSet, ranges, minVals = autoNorm(returnMat)
123
124 # #测试我们的分类器
125 # datingClassTest()
126
127 #请求用户输入几个特征，然后进行分类，判断喜爱程度
128 classifyPerson()
129
130 #draw
131 #drawPoint(normDataSet, classLabelVector)
2.自己的实现
1 # -*- coding: utf-8 -*-
2
3 '''
4 function: 使用二维数据点集(x,y)来实现knn算法核心功能
5 note: 计算欧式距离还可以用 numpy.linalg.norm(array1 - array2) 必须是array
6 date: 2017.07.23
7 '''
8 from numpy import *
9 from random import *
10 import matplotlib.pyplot as plt
11
12 #创建用来训练的数据集和用来测试的数据集，及标注信息,先用100个点训练,20个点测试
13 def createData():
14 trainingList0 = [[(1.0,1.0) for j in range(1)] for i in range(50)]
15 trainingList1 = [[(1.0,1.0) for j in range(1)] for i in range(50)]
16 trainingLabel,trainingLabe0 = [], []
17 for i in range(50):
18 x = randint(1,10) #产生1，10之间的随机浮点数，类别0
19 y = randint(1,10)
20 trainingList0[i] = [(x,y)]
21 trainingLabel.append(0) #属于类别0
22 for j in range(50):
23 x = randint(10,20) #产生十位数，10-20，类别1
24 y = randint(10,20)
25 trainingList1[j] = [(x,y)]
26 trainingLabel.append(1) #属于类别1
27 trainingList = trainingList0 + trainingList1
28
29 #产生测试的数据集
30 testList0 = [[(1.0,1.0) for j in range(1)] for i in range(10)]
31 testList1 = [[(1.0,1.0) for j in range(1)] for i in range(10)]
32 testLabel = []
33 for i in range(10):
34 x = randint(4,9)
35 y = randint(2,10)
36 testList0[i] = [(x,y)]
37 testLabel.append(0)
38 for j in range(10):
39 x = randint(11,19)
40 y = randint(10,18)
41 testList1[j] = [(x,y)]
42 testLabel.append(1)
43 testList = testList0 + testList1
44 print(trainingList)
45 return trainingList, trainingLabel, testList, testLabel
46
47 #对测试数据集和训练数据集之间的距离进行计算，采用欧式距离 d = ((x1-x2)^2 + (y1-y2)^2)^0.5
48 def calculateKNN(trainingList, trainingLabel, testList, testLabel, k):
49 #20行，100列距离，每行代表一个测试点距离100个训练点的距离,初始化为0
50 d = [[0 for j in range(100)] for i in range(20)]
51 #kPointDisList = [1000] * k #初始化k个最近距离为1000
52 for i in range(20):
53 for j in range(100):
54 d[i][j] = getDistance(testList[i], trainingList[j]) #计算距离
55 #这里判断求得的距离是否是最小的K个之一，若是则记录下j,更新此测试点的标签类别
56 #kPointDisList = updateKPoint(d[i][j], kPointDisList)
57 testLable = getKMin(d,k,trainingLabel)
58 print(testLabel)
59 return testLabel
60
61 #计算列表里面的K个最小值，找出对应的标签类别号
62 def getKMin(dList,k,trainingLabel):
63 testLabel = []
64 sortedIndexList = argsort(dList) #返回dLlist从小到大排列的索引的数组
65 print(sortedIndexList)
66 print(len(sortedIndexList))
67 for x in range(20):
68 type1 = 0
69 type0 = 0
70 for i in range(k): #计算最近的K个点，按照类别的多少进行投票表决
71 if trainingLabel[sortedIndexList[x][i]] == 1:
72 type1 += 1
73 else:
74 type0 += 1
75 if type1 > type0:
76 testLabel.append(1)
77 else:
78 testLabel.append(0)
79 return testLabel
80
81 #跟新最近的K个点坐标，标签
82 def updateKPoint(d, kPointDisList):
83 if d < max(kPointDisList):
84 kPointDisList[kPointDisList.index(max(kPointDisList))] = d
85 else:
86 pass
87 return kPointDisList
88
89 #计算两个点的欧式距离,因为有float所以只能用**
90 def getDistance(a, b):
91 return ((a[0][0] - b[0][0])**2 + (a[0][1] - b[0][1])**2)**0.5
92
93 #计算KNN分类结果的正确率
94 def getCorrectRate(testLabel, resultLabel):
95 correctNum = 0
96 for i in range(len(testLabel)):
97 if testLabel[i] == resultLabel[i]:
98 correctNum += 1
99 return correctNum / float(len(testLabel))
100
101 #把分类结果用图形画出来
102 def drawKNN(trainingList, testList):
103 #产生画图数据
104 x1,x2,y1,y2 = [],[],[],[]
105 for i in range(len(trainingList)):
106 x1.append(trainingList[i][0][0])
107 y1.append(trainingList[i][0][1])
108 for i in range(len(testList)):
109 x2.append(testList[i][0][0])
110 y2.append(testList[i][0][1])
111 #创建一个绘图对象
112 fig = plt.figure()
113 ax1 = fig.add_subplot(111)
114 #添加坐标图修饰
115 plt.xlabel('times(m)')
116 plt.ylabel('money(y)')
117 plt.title('first pic')
118 #画散点图,maker代表散点图形状，c代表颜色
119 pl1 = ax1.scatter(x1,y1,c = 'r',marker = '.')
120 pl2 = ax1.scatter(x2,y2,c = 'b',marker = '+')
121 #设置图标
122 plt.legend([pl1,pl2],('train','test'))
123 #显示图形
124 plt.show()
125 #保存图形到本地
126 plt.savefig('test.png')
127 #系统的总体控制逻辑
128 def testKNN():
129 #设定系统的K值
130 k = 3
131 trainingList, trainingLabel, testList, testLabel = createData()
132 resultLabel = calculateKNN(trainingList, trainingLabel, testList, testLabel, k)
133 correctRate = getCorrectRate(testLabel, resultLabel)
134 print('correctRate is = ' + str(correctRate))
135 drawKNN(trainingList, testList)
136
137 testKNN()
138 #drawKNN()