二分K-means的python实现

前两天无意间看到”zouxy09“大牛的机器学习算法与Python实践之（六）二分k均值聚类讲解，我感觉很受启发啦，但是又看到下面的评论里说出了一些不足，然后就抱着试一试的心态去做了一下，所以数据还是用的”zouxy09“的，链接在下面贴出来了。下面放代码啦~

#coding:utf-8
#二分K-means算法
import numpy as np
import random
import math
import matplotlib.pyplot as plt

def readText(inputfile):
lines = open(inputfile,"r").readlines()
dataSet = []
for line in lines:
dataList = line.strip().split()
dataSet.append([float(dataList[0]),float(dataList[1])])
return dataSet

def transDataSet(dataSet):
hang,lie = dataSet.shape
new_data = np.zeros((hang,lie))
for i in range(hang):
new_data[i,:] = dataSet[i,:]
return new_data

def initClusterpoints(dataSet,k):
hang,lie = dataSet.shape
clusterPoints = np.zeros((k,lie))
for i in range(k):
index = int(random.uniform(0,hang))
clusterPoints[i,:] = dataSet[index,:]
return clusterPoints

def calDis(v1,v2):
s = (sum(pow(v1-v2,2)))
return s

def calSSE(dataSubset,cleaterPoint):
dataSubset = transDataSet(dataSubset)
cleaterPoint = transDataSet(cleaterPoint)
dataSubset_hang = dataSubset.shape[0]
sum_dis = 0
for i in range(dataSubset_hang):
sum_dis = sum_dis +calDis(dataSubset[i,:],cleaterPoint[0,:])
return sum_dis

def bisKmeans(dataSet,k):
# 将初始的一个cluster一分为二
hang,lie = dataSet.shape
# clusterCenter = np.zeros((k,lie))
clusterAssment = np.zeros((hang,lie))
for i in range(hang):
clusterAssment[i,0] = 0
currentClusterPoints = np.mean(dataSet,axis=0).tolist()[0]
cenList = {0:[currentClusterPoints]}
clusterNum = 1

while clusterNum < k:
maxSSE = 0.00001
maxSSEIndex = 0.0
for j in range(clusterNum):#得到最大的SSE
currentDataSet = dataSet[np.nonzero(clusterAssment[:,0]==j)[0]]
currentSSE = calSSE(currentDataSet,np.mat(cenList[j]))

if currentSSE >= maxSSE:
maxSSE = currentSSE
maxSSEIndex = j

currentDataSet = dataSet[np.nonzero(clusterAssment[:,0]==maxSSEIndex)[0]]
currentClusterPoints,currentClusterAssment = Kmeans(currentDataSet,2)

if clusterNum ==1:
clusterAssment = currentClusterAssment

else:
#把新分出来的两部分分别打上标签。
currentClusterAssment[np.nonzero(currentClusterAssment[:,0]==1)[0],0]=clusterNum
currentClusterAssment[np.nonzero(currentClusterAssment[:,0]==0)[0],0]=maxSSEIndex

clusterAssment[np.nonzero(clusterAssment[:,0]==maxSSEIndex)[0],:]=currentClusterAssment

#更新聚类中心
cenList[clusterNum] = currentClusterPoints[1,:]
cenList[maxSSEIndex] = currentClusterPoints[0,:]

clusterNum += 1

return cenList,clusterAssment

def Kmeans(dataSet,k):
clusterPoints = initClusterpoints(dataSet,k)
new_dataSet = transDataSet(dataSet)
hang,lie= dataSet.shape
clusterAssment = np.zeros((hang,lie))
clusterChanged = True

while clusterChanged:
clusterChanged = False
#计算每个点与各个中心点的距离
for i in range(hang):
min_dis = 10000.0
min_index = 0
for j in range(k):
distance = calDis(new_dataSet[i,:],clusterPoints[j,:])
if distance < min_dis:
min_dis = distance
min_index = j
#将每个点归类
if clusterAssment[i,0] != min_index:
clusterChanged = True
clusterAssment[i,0] = min_index
#更新聚类中心
for j in range(k):
points = dataSet[np.nonzero(clusterAssment[:,0]==j)[0]]
clusterPoints[j,:] = np.mean(points,axis=0)
# print(u"聚类成功！")
return clusterPoints,clusterAssment

def showCluster(dataSet, k, centroids, clusterAssment):
plt.rcParams["font.sans-serif"]=["SimHei"]
plt.rcParams["axes.unicode_minus"]=False
numSamples, dim = dataSet.shape
plt.title(u"K值为%d的聚类结果"%k)
if dim != 2:
print ("Sorry! I can not draw because the dimension of your data is not 2!")
return 1

mark = ['or', 'ob', 'og', 'ok', '^r', '+r', 'sr', 'dr', '<r', 'pr']
if k > len(mark):
print ("Sorry! Your k is too large! please contact Zouxy")
return 1

# draw all samples
for i in range(numSamples):
markIndex = int(clusterAssment[i, 0])
plt.plot(dataSet[i, 0], dataSet[i, 1], mark[markIndex])#把所有点先画出来，根据聚类结果来标记不同分类的点。

mark = ['Dr', 'Db', 'Dg', 'Dk', '^b', '+b', 'sb', 'db', '<b', 'pb']
# draw the centroids
for i in range(k):#画出聚类中心
plt.plot(centroids[i][0], centroids[i][1], mark[i], markersize = 12)
plt.show()

if __name__=="__main__":
inputfile = r"E:\MyPython\Code\JustTest\testSet.txt"
dataSet = np.mat(readText(inputfile))
a,b=bisKmeans(dataSet,4)
showCluster(dataSet, 4, a, b)