机器学习——kNN（2）示例：改进约会网站的配对效果

=================================版权声明=================================

版权声明：原创文章 禁止转载 

请通过右侧公告中的“联系邮箱（wlsandwho@foxmail.com）”联系我

勿用于学术性引用。

勿用于商业出版、商业印刷、商业引用以及其他商业用途。　　　　　　　　　　　　　　　　

本文不定期修正完善。

本文链接：http://www.cnblogs.com/wlsandwho/p/7587203.html

耻辱墙：http://www.cnblogs.com/wlsandwho/p/4206472.html

=======================================================================

这个示例实际上是对kNN的练习，区别是使用来自文件的数据。

=======================================================================

1从文件中读取数据并格式化为指定方式。

文件为file2matrix.py

 from numpy import *
 '''
 def file2matrix1(filename):
     f=open(filename)
     arrlines=f.readlines()
     rows=len(arrlines)
     retmat=zeros((rows,3))
     vctlabels=[]
     index=0

     for line in arrlines:
         line=line.strip()
         list=line.split("\t")
         retmat[index,0:3]=list[0:3]
         vctlabels.append(int(list[-1]))
         index+=1

     return retmat,vctlabels

 mat1,labels1=file2matrix1("datingTestSet2.txt")
 print(mat1)
 print(labels1)
 '''
 def file2matrix(filename):
     with open(filename) as file:
         line1=file.readline()
         list1=line1.split()
         cols=len(list1)
         file.seek(0,0)
         lines=file.readlines()
         rows=len(lines)

         index=0
         labels=[]
         realcol=cols-1
         retmat=zeros((rows,realcol))
         for line in lines:
             list=line.split()
             retmat[index,:]=list[0:realcol]
             labels.append(int(list[-1]))

             index+=1
     return retmat,labels

 def file2matrix2(filename):
     with open(filename) as file:
         line1=file.readline()
         list1=line1.split()
         cols=len(list1)
         file.seek(0,0)
         lines=file.readlines()
         rows=len(lines)

         index=0
         labels=[]
         realcol=cols-1
         retmat=zeros((rows,realcol))
         for line in lines:
             list=line.split()
             retmat[index,:]=list[0:realcol]
             #labels.append(int(list[-1]))
             if("largeDoses"==list[-1]):
                 labels.append(3)
             elif("smallDoses"==list[-1]):
                 labels.append(2)
             elif("didntLike"==list[-1]):
                 labels.append(1)
             index+=1
     return retmat,labels

测试一下

文件为test_file2matrix.py

 from file2matrix import *

 mat,labels=file2matrix("datingTestSet2.txt")
 print(mat)
 print(labels)

结果（红框是因为我用的虚拟机，不要在意这些细节）

=======================================================================

2 做个图看看相关性/趋势。实际上这里并没有做数学上的讨论，就是画个图看看臆测一下。

文件是drawapicture1.py

 import matplotlib
 import matplotlib.pyplot as plt
 import numpy
 from file2matrix import *
 mat,labels=file2matrix("datingTestSet2.txt")
 nSizeofLabels=len(labels)

 fig=plt.figure()
 ax=fig.add_subplot(111)
 ax.scatter(mat[:,1],mat[:,2],s=15.0*array(labels),c=15.0*array(labels))
 plt.show()

结果

文件是drawapicture2.py

 import matplotlib
 import matplotlib.pyplot as plt
 import numpy
 from file2matrix import *

 mat,labels=file2matrix("datingTestSet2.txt")
 nSizeofLabels=len(labels)
 mat1x=[]
 mat2x=[]
 mat3x=[]
 mat1y=[]
 mat2y=[]
 mat3y=[]
 for i in range(nSizeofLabels):
     if  labels[i]==1:
         mat1x.append(mat[i][0])
         mat1y.append(mat[i][1])
     elif labels[i]==2:
         mat2x.append(mat[i][0])
         mat2y.append(mat[i][1])
     elif  labels[i]==3:
         mat3x.append(mat[i][0])
         mat3y.append(mat[i][1])

 fig=plt.figure()
 ax=fig.add_subplot(111)
 lg1=ax.scatter(mat1x,mat1y,s=20,c='red')
 lg2=ax.scatter(mat2x,mat2y,s=20,c='green')
 lg3=ax.scatter(mat3x,mat3y,s=20,c='blue')
 fig.legend((lg1,lg2,lg3),('yiban','xihuan','milian'),"upper left")
 plt.show()

结果是

=======================================================================

3 由于不同意义的数据的取值范围很大，所以需要归一化。

文件是matrixnormalization.py

 from  numpy import *

 def autonorm(mat):
     minv=mat.min(0)
     maxv=mat.max(0)
     diff=maxv-minv
     rows=mat.shape[0]
     normmat=zeros(shape(mat))
     normmat=mat-tile(minv,(rows,1))
     normmat=normmat/tile(diff,(rows,1))
     return normmat,minv,maxv,diff

测试归一化

文件是test_matrixnormalization.py

 from matrixnormalization import *

 mat=array([[1,20,3000],[5,60,7000],[2,30,3],[6,60,6000]])
 normmat=autonorm(mat)
 print(normmat)

结果（不知道为何，跟书上的结果不太一样。）

=======================================================================

4测试一下这个分类器的效果

文件名test_dating.py

 from file2matrix import *
 from matrixnormalization import *
 from kNN import *

 def testdating():
     ratio=0.1
     countforerr=0

     mat,lab=file2matrix2("datingTestSet.txt")

     normmat,minv,maxv,diff=autonorm(mat)

     allrows=normmat.shape[0]
     rowsforTest=int(allrows*ratio)

     for i in range(rowsforTest):
         res=classify_kNN(normmat[i,:],normmat[rowsforTest:allrows,:],lab[rowsforTest:allrows],3)
         print('the result is',res,'the real is',lab[i])
         if (res!=lab[i]):
             countforerr+=1.0

     print("the error rate is", (countforerr/float(allrows)))

 testdating()

结果

=======================================================================

5支持手工输入

文件名test_dating2.py

 from file2matrix import *
 from matrixnormalization import *
 from kNN import *

 def testdating():
     ratio=0.1
     countforerr=0
     reslist=["not at all","a little","very well"]
     mat,lab=file2matrix2("datingTestSet.txt")

     normmat,minv,maxv,diff=autonorm(mat)

     allrows=normmat.shape[0]
     rowsforTest=int(allrows*ratio)

     percenttats=float(input("percentage of time spent playing video games?"))
     flymiles=float(input("frequent filer miles earned per year?"))
     icecream=float(input("liters of ice cream consumed per year?"))
     desarr=array([flymiles,percenttats,icecream])
     normdesarr=(desarr-minv)/diff

     res=classify_kNN(normdesarr,normmat[rowsforTest:allrows,:],lab[rowsforTest:allrows],3)
     print("You will probably like this person:",reslist[res-1])

 testdating()

结果

=======================================================================

这一节更像是用一个小算法领着学习python。