import numpy as np
import sklearn.datasets             #加載原數據
import matplotlib.pyplot as plt
import random

#點到各點距離
def PointToData(point,dataset):
    a = np.multiply(dataset - point,dataset - point)
    # print(‘a‘,a)
    distence = np.sqrt(a[:,0]+a[:,1])
    return distence

#選擇初始的k個中心簇
 

def startpoint(k,dataset):
    m, n = np.shape(dataset)
    index1 = random.randint(0,len(dataset) - 1)
    A = []  # 初始的k個中心簇
    A_dit = []  # 初始所有點到中心簇的距離
    A.append(dataset[index1])
    sum_dis = np.zeros((m, 1))
    flag_mat = np.ones((m,1))
    flag_mat[index1] = 0
    for i in range(0, k - 1):
        A_dit.append((PointToData(A[i], dataset)).reshape(
 
-1,1) )
        # print(‘A_dit[{}]:{}‘.format(i,A_dit[i]))
        sum_dis =(sum_dis  + A_dit[i]) * flag_mat
        # print(‘sum_dis[{}]:{}‘.format(i,sum_dis))
        Index = np.argmax(sum_dis)
        flag_mat[Index] = 0
        # print(‘選的Index：‘,Index)
        A.append(dataset[Index])
    
 
return A

#加載數據
Data = sklearn.datasets.load_iris()
dataset = Data.data[:,0:2]

# #小數據測試編碼
# test = dataset[0:15,:]
# testm,testn = np.shape(test)
# print(test)

#測試k
# k = 4
#初始點測試函數
# Apoint = startpoint(k,test)
# print(‘Apoint‘,Apoint)
#距離函數測試
# d = PointToData(test[0,:],test)
# print(‘d,d+d:‘,d,d+d)

def classfy(dataset,Apoint):
    m,n = np.shape(dataset)
    dis_li = []
    num = 0
    for point in Apoint:
        distence = PointToData(point,dataset)
        dis_li.append(distence)
        if num == 0:
            dis_li_mat = dis_li[num]
        else:
            dis_li_mat = np.column_stack((dis_li_mat,dis_li[num]))
        num += 1
    result = np.argmin(dis_li_mat,axis=1)
    # print(‘dis_li:‘,dis_li)
    # print(‘dis_li_mat:\n‘, dis_li_mat)
    # print(‘classfy:‘,result)
    return result
# label2 = classfy(test,Apoint)
# print(‘label2:‘,label2)

#求分類的新中心
def Center(dataset,label,k):
    i = 0
    newpoint = []
    for index in range(k):
        flag = (label==index)
        # print(‘flag,i:‘,flag,i)
        num = sum(flag)
        # print(‘num:‘,num,index)
        a = flag.reshape(-1,1) * dataset
        newpoint.append(np.sum(a,axis = 0)/num)
        i += 1
    # print(newpoint)
    return newpoint
# testcenter = center(test,label2,k)
# print(‘testcenter:‘,testcenter)

#K-means主體函數
def myK(k,dataset):
    Startpoint = startpoint(k,dataset)
    m,n = np.shape(Startpoint)
    centerpoint = Startpoint
    labelset = classfy(dataset,Startpoint)
    newcenter = Center(dataset,labelset,k)
    # print(‘外:cecnterpoint‘, centerpoint)
    # print(‘外:newcenter‘, newcenter)
    flag = 0
    for i in range(k):
        for j in range(n):
            if centerpoint[i][j] != newcenter[i][j]:
                flag = 1
    while flag:
        print(‘循環‘)
        # print(‘裏:cecnterpoint‘, centerpoint)
        # print(‘裏:newcenter‘, newcenter)
        flag = 0
        for i in range(k):
            for j in range(n):
                if centerpoint[i][j] != newcenter[i][j]:
                    flag = 1
        # print(‘flag:‘,flag)
        centerpoint = newcenter[:]
        labelset = classfy(dataset,centerpoint)
        newcenter = Center(dataset, labelset, k)
    # print(‘final_resultlabel:‘,labelset)
    # print(‘cenerpoint:‘, centerpoint)
    return labelset,centerpoint

#測試
k=5
final_label,centerpoint = myK(k,dataset)
print(‘centerpoint:‘,centerpoint)
mat_center = np.mat(centerpoint)

#畫圖
# plt.scatter(test[:,0],test[:,1],40,10*(labelset+1))
plt.scatter(dataset[:, 0], dataset[:, 1],40,10*(final_label+1))
plt.show()

機器學習之--kmeans聚類簡單算法實例