k近邻法( k-nearnest neighbor)

基本思想：

给定一个训练数据集，对新的输入实例，在训练数据集中找到与该实例最邻近的k个实例，这k个实例的多数属于某个类，就把该输入实例分为这个类

---

距离度量：

特征空间中两个实例点的距离是两个实例点相似程度的反映，一般常用欧氏距离，更一般的有行内公式(L_p)或者Minkowski距离

[L_p(x_i,x_j)=(sum_{l=1}^{n}|x_i^{(l)}-x_j^{(l)}|^p)^{frac{1}{p}} ]

• 当(p=1)时，为曼哈顿距离，(L_1(x_i,x_j)=sum_{l=1}^{n}|x_i{(l)}-x_j^{(l)}|)

• 当(p=2)时，为欧式距离，(L_2(x_i,x_j)=(sum|x_i^{(l)}-x_j{(l)}|²⁾{frac{1}{2}})

• 当(p=infty)时，它是各个坐标距离的最大值，(L_infty(x_i,x_j)=max_l|x_i^{(l)}-x_j{(l)}|)

机器学习实战第二章代码

import numpy as np
def classify0(inx,dataSet,labels,k):
    datasize=dataSet.shape[0]
    diffmat=np.tile(inx,(datasize,1))-dataSet
    sqdismat=diffmat**2
    sqdist=sqdismat.sum(axis=1)
    dist=sqdist**0.5
    sortdistpos=dist.argsort()
    labelscount=np.array([0,0,0,0])
    for i in range(k):
        votelabels=labels[sortdistpos[i]]
        labelscount[votelabels]+=1
    returnresult=labelscount.argsort()
    return returnresult[-1]

def dataload(filename):
    file=open(filename)
    ar=file.readlines()
    num=len(ar)
    returnMat=np.zeros((num,3))
    returnLabels=[]
    index=0
    for line in ar:
        line=line.strip()
        linelist=line.split('	')
        returnMat[index,:]=linelist[0:3]
        returnLabels.append(int(linelist[-1]))
        index+=1
    return returnMat,returnLabels

def autoNorm(dataSet):
    minvals=dataSet.min(0)
    maxvals=dataSet.max(0)
    ranges=maxvals-minvals
    normDataSet=np.zeros(np.shape(dataSet))
    m=dataSet.shape[0]
    normDataSet=dataSet-np.tile(minvals,(m,1))
    normDataSet=normDataSet/np.tile(ranges,(m,1))
    return normDataSet

DataMat,DataLabels=dataload('datingTestSet2.txt')
normDataMat=autoNorm(DataMat)
ratio=0.1
tep=normDataMat.shape[0]
testnum=int(tep*ratio)
print(tep,testnum)
errorcount=0
for i in range(testnum):
    result=classify0(normDataMat[i,:],normDataMat[testnum:tep,:],DataLabels[testnum:tep],3)
    if(result!=DataLabels[i]):
        errorcount+=1
    print(i," the test result is ",result,",the real result is ",DataLabels[i])
print(errorcount)
print("the error ratio is ",errorcount*1.0/testnum)