python中kmeans聚类实现代码

k-means算法思想较简单，说的通俗易懂点就是物以类聚，花了一点时间在python中实现k-means算法，k-means算法有本身的缺点，比如说k初始位置的选择，针对这个有不少人提出k-means++算法进行改进；另外一种是要对k大小的选择也没有很完善的理论，针对这个比较经典的理论是轮廓系数，二分聚类的算法确定k的大小，在最后还写了二分聚类算法的实现，代码主要参考机器学习实战那本书：

#encoding:utf-8
'''''
Created on 2015年9月21日
@author: ZHOUMEIXU204
'''

path=u"D:\\Users\\zhoumeixu204\\Desktop\\python语言机器学习\\机器学习实战代码  python\\机器学习实战代码\\machinelearninginaction\\Ch10\\"
import numpy as np
def loadDataSet(fileName): #读取数据
 dataMat=[]
 fr=open(fileName)
 for line in fr.readlines():
   curLine=line.strip().split('\t')
   fltLine=map(float,curLine)
   dataMat.append(fltLine)
 return dataMat
def distEclud(vecA,vecB):  #计算距离
 return np.sqrt(np.sum(np.power(vecA-vecB,2)))
def randCent(dataSet,k):   #构建镞质心
 n=np.shape(dataSet)[1]
 centroids=np.mat(np.zeros((k,n)))
 for j in range(n):
   minJ=np.min(dataSet[:,j])
   rangeJ=float(np.max(dataSet[:,j])-minJ)
   centroids[:,j]=minJ+rangeJ*np.random.rand(k,1)
 return centroids
dataMat=np.mat(loadDataSet(path+'testSet.txt'))
print(dataMat[:,0])

# 所有数都比-inf大
# 所有数都比+inf小
def kMeans(dataSet,k,distMeas=distEclud,createCent=randCent):
 m=np.shape(dataSet)[0]
 clusterAssment=np.mat(np.zeros((m,2)))
 centroids=createCent(dataSet,k)
 clusterChanged=True
 while clusterChanged:
   clusterChanged=False
   for i in range(m):
     minDist=np.inf;minIndex=-1 #np.inf表示无穷大
     for j in range(k):
       distJI=distMeas(centroids[j,:],dataSet[i,:])
       if distJI
         minDist=distJI;minIndex=j
     if clusterAssment[i,0]!=minIndex:clusterChanged=True
     clusterAssment[i,:]=minIndex,minDist**2
   print centroids
   for cent in range(k):
     ptsInClust=dataSet[np.nonzero(clusterAssment[:,0].A==cent)[0]] #[0]这里取0是指去除坐标索引值，结果会有两个
     #np.nonzero函数，寻找非0元素的下标 nz=np.nonzero([1,2,3,0,0,4,0])结果为0,1,2
     centroids[cent,:]=np.mean(ptsInClust,axis=0)

return centroids,clusterAssment
myCentroids,clustAssing=kMeans(dataMat,4)  
print(myCentroids,clustAssing)  

#二分均值聚类(bisecting k-means)
def  biKmeans(dataSet,k,distMeas=distEclud):
 m=np.shape(dataSet)[0]
 clusterAssment=np.mat(np.zeros((m,2)))
 centroid0=np.mean(dataSet,axis=0).tolist()[0]
 centList=[centroid0]
 for j in range(m):
   clusterAssment[j,1]=distMeas(np.mat(centroid0),dataSet[j,:])**2
 while (len(centList)
   lowestSSE=np.Inf
   for i in range(len(centList)):
     ptsInCurrCluster=dataSet[np.nonzero(clusterAssment[:,0].A==i)[0],:]
     centroidMat,splitClusAss=kMeans(ptsInCurrCluster,2,distMeas)
     sseSplit=np.sum(splitClusAss[:,1])
     sseNotSplit=np.sum(clusterAssment[np.nonzero(clusterAssment[:,0].A!=i)[0],1])
     print "sseSplit, and notSplit:",sseSplit,sseNotSplit
     if (sseSplit+sseNotSplit)
       bestCenToSplit=i
       bestNewCents=centroidMat
       bestClustAss=splitClusAss.copy()
       lowestSSE=sseSplit+sseNotSplit
   bestClustAss[np.nonzero(bestClustAss[:,0].A==1)[0],0]=len(centList)
   bestClustAss[np.nonzero(bestClustAss[:,0].A==0)[0],0]=bestCenToSplit
   print "the bestCentToSplit is:",bestCenToSplit
   print 'the len of bestClustAss is:',len(bestClustAss)
   centList[bestCenToSplit]=bestNewCents[0,:]
   centList.append(bestNewCents[1,:])
   clusterAssment[np.nonzero(clusterAssment[:,0].A==bestCenToSplit)[0],:]=bestClustAss
 return centList,clusterAssment
print(u"二分聚类分析结果开始")
dataMat3=np.mat(loadDataSet(path+'testSet2.txt'))
centList,myNewAssments=biKmeans(dataMat3, 3)
print(centList)

来源：http://blog.csdn.net/luoyexuge/article/details/49105177