1 def biKmeans(dataSet, k, distMeas=distEclud):
 2     m = shape(dataSet)[0]
 3     clusterAssment = mat(zeros((m,2)))#記錄簇分配的結果及誤差
 4     centroid0 = mean(dataSet, axis=0).tolist()[0]#計算整個資料集的質心
 5     centList =[centroid0] #create a list with one centroid
 6     for j in range(m):#計算初始聚類點與其他點的距離
 7         clusterAssment[j,1] = distMeas(mat(centroid0), dataSet[j,:])**2
 8
 
     while (len(centList) < k):
 9         lowestSSE = inf
10         for i in range(len(centList)):#嘗試劃分每一簇
11             ptsInCurrCluster = dataSet[nonzero(clusterAssment[:,0].A==i)[0],:]#get the data points currently in cluster i
12             centroidMat, splitClustAss = kMeans(ptsInCurrCluster, 2, distMeas)#
 
對這個簇執行一個KMeans演算法，k=2
13             sseSplit = sum(splitClustAss[:,1])#compare the SSE to the currrent minimum
14             sseNotSplit = sum(clusterAssment[nonzero(clusterAssment[:,0].A!=i)[0],1])
15             print "sseSplit, and notSplit: ",sseSplit,sseNotSplit
16             if (sseSplit + sseNotSplit) < lowestSSE:#
 
#劃分後更好的話
17                 bestCentToSplit = i
18                 bestNewCents = centroidMat
19                 bestClustAss = splitClustAss.copy()
20                 lowestSSE = sseSplit + sseNotSplit
21         bestClustAss[nonzero(bestClustAss[:,0].A == 1)[0],0] = len(centList) #更新簇的分配結果change 1 to 3,4, or whatever
22         bestClustAss[nonzero(bestClustAss[:,0].A == 0)[0],0] = bestCentToSplit
23         print 'the bestCentToSplit is: ',bestCentToSplit
24         print 'the len of bestClustAss is: ', len(bestClustAss)
25         centList[bestCentToSplit] = bestNewCents[0,:].tolist()[0]#replace a centroid with two best centroids 
26         centList.append(bestNewCents[1,:].tolist()[0])
27         clusterAssment[nonzero(clusterAssment[:,0].A == bestCentToSplit)[0],:]= bestClustAss#reassign new clusters, and SSE
28     return mat(centList), clusterAssment