Python实现KMeans

糖逗

参考书籍：《机器学习实战》



1.K均值聚类

1. import numpy as np
   
2. 
   
3. def loadDataSet(fileName):
   
4.     dataMat = []
   
5.     fr = open(fileName)
   
6.     for line in fr.readlines():
   
7.         curLine = line.strip().split("\t")
   
8.         fltLine = list(map(float, curLine))
   
9.         dataMat.append(fltLine)
   
10.     return np.mat(dataMat)
    
11. 
    
12. 
    
13. 
    
14. def distance(vecA, vecB):
    
15.     return np.sqrt(np.sum(np.power(vecA - vecB, 2)))
    
16. 
    
17. 
    
18. #根据每个特征的最大值最小值范围随机生成k个质心
    
19. def randCent(dataSet, k):
    
20.     n = np.shape(dataSet)[1]
    
21.     centroids = np.mat(np.zeros((k, n)))
    
22.     for j in range(n):
    
23.         minJ = min(dataSet[:, j])
    
24.         rangeJ = float(max(dataSet[:, j]) - minJ)
    
25.         #https://blog.csdn.net/sinat_38944746/article/details/89140276
    
26.         centroids[:, j] = minJ + rangeJ * np.random.rand(k, 1)
    
27.     return centroids
    
28. 
    
29. 
    
30. def kMeans(dataSet, k, distMeas = distance, createCent = randCent):
    
31.     m = np.shape(dataSet)[0]
    
32.     clusterAssment = np.mat(np.zeros((m, 2)))#存储每个样本点最近质心的ID和距离
    
33.     centroids = createCent(dataSet, k)
    
34.     clusterChanged = True
    
35.     while clusterChanged:
    
36.         clusterChanged = False
    
37.         for i in range(m):
    
38.             minDist = np.inf
    
39.             minIndex = -1
    
40.             for j in range(k):#对于每个样本点找其最近的质心
    
41.                 distJI = distMeas(centroids[j, :], dataSet[i, :])
    
42.                 if distJI < minDist:
    
43.                     minDist = distJI
    
44.                     minIndex = j
    
45.             if clusterAssment[i, 0] != minIndex:
    
46.                 clusterChanged = True#该样本的最近质心改变了
    
47.             clusterAssment[i, :] = minIndex, minDist ** 2
    
48.         for cent in range(k):#重新计算质心
    
49.             ptsInClust = dataSet[np.nonzero(clusterAssment[:, 0] == cent)[0]]
    
50.             centroids[cent, :] = np.mean(ptsInClust, axis = 0)
    
51.     return centroids, clusterAssment
    
52.    
    
53. 
    
54. if __name__ == "__main__":
    
55.     dataMat = loadDataSet(r"C:\...\testSet.txt")
    
56.     myCentroids, clustAssing = kMeans(dataMat, 4)

复制代码

2.二分K均值聚类

1. import numpy as np
   
2. 
   
3. def loadDataSet(fileName):
   
4.     dataMat = []
   
5.     fr = open(fileName)
   
6.     for line in fr.readlines():
   
7.         curLine = line.strip().split("\t")
   
8.         fltLine = list(map(float, curLine))
   
9.         dataMat.append(fltLine)
   
10.     return np.mat(dataMat)
    
11. 
    
12. 
    
13. 
    
14. def distance(vecA, vecB):
    
15.     return np.sqrt(np.sum(np.power(vecA - vecB, 2)))
    
16. 
    
17. 
    
18. #根据每个特征的最大值最小值范围随机生成k个质心
    
19. def randCent(dataSet, k):
    
20.     n = np.shape(dataSet)[1]
    
21.     centroids = np.mat(np.zeros((k, n)))
    
22.     for j in range(n):
    
23.         minJ = min(dataSet[:, j])
    
24.         rangeJ = float(max(dataSet[:, j]) - minJ)
    
25.         #https://blog.csdn.net/sinat_38944746/article/details/89140276
    
26.         centroids[:, j] = minJ + rangeJ * np.random.rand(k, 1)
    
27.     return centroids
    
28. 
    
29. def kMeans(dataSet, k, distMeas = distance, createCent = randCent):
    
30.     m = np.shape(dataSet)[0]
    
31.     clusterAssment = np.mat(np.zeros((m, 2)))#存储每个样本点最近质心的ID和距离
    
32.     centroids = createCent(dataSet, k)
    
33.     clusterChanged = True
    
34.     while clusterChanged:
    
35.         clusterChanged = False
    
36.         for i in range(m):
    
37.             minDist = np.inf
    
38.             minIndex = -1
    
39.             for j in range(k):#对于每个样本点找其最近的质心
    
40.                 distJI = distMeas(centroids[j, :], dataSet[i, :])
    
41.                 if distJI < minDist:
    
42.                     minDist = distJI
    
43.                     minIndex = j
    
44.             if clusterAssment[i, 0] != minIndex:
    
45.                 clusterChanged = True#该样本的最近质心改变了
    
46.             clusterAssment[i, :] = minIndex, minDist ** 2
    
47.         for cent in range(k):#重新计算质心
    
48.             ptsInClust = dataSet[np.nonzero(clusterAssment[:, 0] == cent)[0]]
    
49.             centroids[cent, :] = np.mean(ptsInClust, axis = 0)
    
50.     return centroids, clusterAssment
    
51. 
    
52. def biKmeans(dataSet, k, distMeas = distance):
    
53.     m = np.shape(dataSet)[0]
    
54.     clusterAssment = np.mat(np.zeros((m, 2)))
    
55.     centroid0 = np.mean(dataSet, axis = 0).tolist()[0]#一开始将虽有样本都聚成一个类，质心就是各特征的均值中心
    
56.     centList =[centroid0]
    
57.     for j in range(m):
    
58.         clusterAssment[j,1] = distMeas(np.mat(centroid0), dataSet[j,:])**2
    
59.     while (len(centList) < k):#当分类达到k类后停止
    
60.         lowestSSE = np.inf
    
61.         for i in range(len(centList)):
    
62.             ptsInCurrCluster = dataSet[np.nonzero(clusterAssment[:,0] == i)[0],:]
    
63.             centroidMat, splitClustAss = kMeans(ptsInCurrCluster, 2, distMeas)
    
64.             sseSplit = sum(splitClustAss[:,1])#将当前类进一步划分为两类后的sse
    
65.             sseNotSplit = sum(clusterAssment[np.nonzero(clusterAssment[:,0] != i)[0],1])
    
66.             #print("sseSplit, and notSplit: ",sseSplit,sseNotSplit)
    
67.             if (sseSplit + sseNotSplit) < lowestSSE:#这里的SSE为什么是换分前和划分后的总sse，为什么不是差值看变化多少呢？
    
68.                 bestCentToSplit = i
    
69.                 bestNewCents = centroidMat
    
70.                 bestClustAss = splitClustAss.copy()
    
71.                 lowestSSE = sseSplit + sseNotSplit
    
72.         bestClustAss[np.nonzero(bestClustAss[:, 0] == 1)[0],0] = len(centList)#给样本重新标记质心
    
73.         bestClustAss[np.nonzero(bestClustAss[:, 0] == 0)[0],0] = bestCentToSplit
    
74.         #print('the bestCentToSplit is: ', bestCentToSplit)
    
75.         #print('the len of bestClustAss is: ', len(bestClustAss))
    
76.         centList[bestCentToSplit] = bestNewCents[0, :].tolist()[0]#修改原质心
    
77.         centList.append(bestNewCents[1, :].tolist()[0])#添加新置信
    
78.         clusterAssment[np.nonzero(clusterAssment[:, 0] == bestCentToSplit)[0], :]= bestClustAss
    
79.     return np.mat(centList), clusterAssment
    
80.    
    
81. 
    
82. if __name__ == "__main__":
    
83.     dataMat = loadDataSet(r"C:\...\testSet2.txt")
    
84.     myCentroids, clustAssing = biKmeans(dataMat, 3)
    
85.     import matplotlib.pyplot as plt
    
86.     plt.scatter(dataMat[:, 0].tolist(), dataMat[:, 1].tolist())
    
87.     plt.plot(myCentroids[:, 0].tolist(), myCentroids[:, 1].tolist(),'rs')

复制代码

糖逗