利用K-均值聚类算法对未标注数据分组

kMean.py

#!/usr/bin/python  
# -*- coding: utf-8 -*-  
#coding=utf-8

from numpy import *

#导入数据
def loadDataSet(fileName):
    dataMat = []
    fr = open(fileName)
    for line in fr.readlines():
        curLine = line.strip().split('\t')
        fltLine = map(float, curLine)
        dataMat.append(fltLine)
    return dataMat

#计算两个向量的欧氏距离
def distEclud(vecA, vecB):
    return sqrt(sum(power(vecA-vecB, 2)))

#为给定数据集构建一个包含k个随机质心的集合
def randCent(dataSet, k):
    n = shape(dataSet)[1]
    centroids = mat(zeros((k, n)))
    for j in range(n):
        minJ = min(dataSet[:, j])
        randJ = float(max(dataSet[:,j] - minJ))
        #random.rand(k, 1)生成k个[0,1]中的随机数
        centroids[:, j] = minJ + randJ * random.rand(k, 1)
    return centroids

#k均值聚类算法
def kMeans(dataSet, k, distMeas=distEclud, createCent=randCent):
    m = shape(dataSet)[0]  #数据点总数
    clusterAssment = mat(zeros((m, 2)))  #簇分配结果矩阵，一列记录簇索引值，一列存储误差
    centroids =createCent(dataSet, k)
    clusterChanged = True
    #按 计算质心-分配-重新计算 反复迭代
    while clusterChanged:
        clusterChanged = False
        for i in range(m):
            minDist = inf
            minIndex = -1
            #寻找最近的质心
            for j in range(k):
                distJI = distMeas(centroids[j,:], dataSet[i,:])
                if distJI < minDist:
                    minDist = distJI
                    minIndex = j
            if clusterAssment[i,0] != minIndex:
                clusterChanged = True
            clusterAssment[i,:] = minIndex, minDist**2
        #更新质心的位置
        for cent in range(k):
            ptsInClust = dataSet[nonzero(clusterAssment[:,0].A==cent)[0]]
            centroids[cent,:] = mean(ptsInClust, axis=0)
    return centroids, clusterAssment

#二分k-均值聚类算法
def biKeans(dataSet, k, distMeas=distEclud):
    m = shape(dataSet)[0]
    clusterAssment = mat(zeros((m, 2)))  #第一例存储簇分配结果，第二列存储平方误差
    centroid0 = mean(dataSet, axis=0).tolist()[0]  #创建一个初始簇
    cenList = [centroid0]  #存储所有质心
    for j in range(m):
        clusterAssment[j,1] = distMeas(mat(centroid0), dataSet[j,:]) ** 2
    while (len(cenList) < k):
        lowestSSE = inf
        for i in range(len(cenList)):  
            ptsInCurrCluster = dataSet[nonzero(clusterAssment[:,0].A==i)[0],:] #将簇中的所有点看成一个小的数据集
            centroidMat, splitClustAss = kMeans(ptsInCurrCluster, 2, distMeas) #生成2个质心簇
            sseSplit = sum(splitClustAss[:,1])  #误差和
            sseNotSplit = sum(clusterAssment[nonzero(clusterAssment[:,0].A != i)[0], 1]) #剩余数据集误差和
            if (sseSplit + sseNotSplit) < lowestSSE:
                bestCentToSplit = i
                bestNewCents = centroidMat
                bestClusAss = splitClustAss.copy()
                lowestSSE = sseSplit + sseNotSplit
        #将要划分的簇中所有点的簇分配结果进行修改
        bestClusAss[nonzero(bestClusAss[:,0].A==0)[0], 0] = bestCentToSplit #将编号为0的簇改为划分簇的编号
        bestClusAss[nonzero(bestClusAss[:,0].A==1)[0], 0] = len(cenList)    #将编号为1的簇改为添加簇的编号
        print 'the bestCentToSplit is: ', bestCentToSplit
        print 'the len of bestClustAss is: ', len(bestClusAss)
        cenList[bestCentToSplit] = bestNewCents[0,:].tolist()[0] #修改质心列表 
        cenList.append(bestNewCents[1,:].tolist()[0])  #添加到质心列表
        clusterAssment[nonzero(clusterAssment[:,0].A == bestCentToSplit)[0],:] = bestClusAss
    return mat(cenList), clusterAssment       

测试

>>> import kMean
>>> dataMat = mat(loadDataSet('testSet2.txt'))
>>> centList, myAssments = biKeans(dataMat, 3)
the bestCentToSplit is:  0
the len of bestClustAss is:  60
the bestCentToSplit is:  0
the len of bestClustAss is:  40
>>> centList
matrix([[-2.94737575,  3.3263781 ],
        [-0.45965615, -2.7782156 ],
        [ 2.93386365,  3.12782785]])
>>> myAssments
matrix([[  2.00000000e+00,   1.45461050e-01],
        [  0.00000000e+00,   6.80213825e-01],
        [  1.00000000e+00,   1.02184582e+00],
        [  2.00000000e+00,   1.34548760e+00],
        [  0.00000000e+00,   1.35376464e+00],
        [  1.00000000e+00,   3.87167519e+00],
        [  2.00000000e+00,   8.37259951e-01],
        [  0.00000000e+00,   2.20116272e-01],
        ... ...