《机器学习实战》第十章

import numpy as np

# read data

defloadData(filename):

    fr = open(filename)

    dataMat = []

for line in fr.readlines():	readlines readline 是读下一行 readlines 是遍历所有行
data1 = line.strip().split('\t')	Strip  split  strip是删除 split 是分离
data = list(map(float,data1))	要加list

       

        dataMat.append(data)

return np.mat(dataMat)

最后要变成矩阵的形式

 

# cal distance

defdisEclud(vecA,vecB):

    returnnp.sqrt(np.sum(np.power(vecA-vecB,2)))

 

# choose center

defrandCent(data,k):

    n = np.shape(data)[1]

cent = np.mat(np.zeros((k,n)))

构建空矩阵要用元祖形式，也就是要俩个括号

    for i in range(n):

        minD = np.min(data[:,i])

rangeD = np.float(np.max(data[:,i]-minD))

？？

        cent[:,i] = minD +rangeD*np.random.rand(k,1)

    return cent

## test

#data =loadData('G:/anaconda4/machinelearninginaction/Ch10/testSet.txt')

#print(data)

#result =randCent(data,3)

#print(result)

 

defKmeans(data,k,distMeas = disEclud,createCent = randCent):

    m = np.shape(data)[0]

    clusterAssment = np.mat(np.zeros((m,2)))

    cent = randCent(data,k)

    clusterChanged = True

    while clusterChanged:

        clusterChanged = False

        for i in range(m):

            minDist = np.inf

            minIndex = -1

            for j in range(k):

                distJ =distMeas(cent[j,:],data[i,:])

                if distJ < minDist:

                    minDist = distJ

                    minIndex = j

            if clusterAssment[i,0] != minIndex:

                clusterChanged = True

clusterAssment[i,:] = minIndex,minDist**2

这边为什么要平方 不是很理解》？？

        print(cent)

        for ii in range(np.shape(cent)[0]):

a = clusterAssment[:,0]             a1 = clusterAssment[:,0].A

加了.A好像没有很大的区别，之前 看说明是说.A是为了转成array的格式

           

b = clusterAssment[:,0] == ii  b1 = clusterAssment[:,0].A == ii

判断是否相等，出来的结果是布尔型数值True False

           

c = np.nonzero(b)	找出非零元素的位置       多维数据返回的结果是二维的，一个维度是行，一个维度是列 如果是bool型的就是返回True 的位置
d = c[0]	返回True所在的行
e = data[d]	True（也就是对应的第k个簇）内包含的元素

Clust = data[np.nonzero(clusterAssment[:,0].A == ii)[0]]

之上的这些abcde都是测试行

            cent[ii,:] = np.mean(Clust,axis =0)          

    return cent,clusterAssment

           

data =loadData('G:/anaconda4/machinelearninginaction/Ch10/testSet.txt')          

Kmeans(data,4,distMeas= disEclud,createCent = randCent)           

   

    综上 就是把每一个簇的对应数组找出来，然后进行求均值