kmean算法python实现

<span style="font-family: Arial, Helvetica, sans-serif;"># -*- coding:utf-8 -*_</span>

import numpy as np
import matplotlib.pyplot as plt
import time

time1=time.time()
class Data:
    def __init__(self,vector,set_id):
        self._vector=vector
        self._set=set_id

class Store:
    def __init__(self):
        self.list=[]
        self.list.append([])
    def add(self,data):
        self.list[0].append(data)
    def num_of_store(self):
        return len(self.list[0])




#聚类数
k=4
store=Store()
file=open('data','r')
line=file.readline()
i=0
inputs=[]
while line:
    string=line.strip('\n').split('\t')
    inputs.append([float(string[0]),float(string[1])])
    line=file.readline()

for point in inputs:
    store.add(Data(point,-1))


while len(store.list)<k+1:
    store.list.append([])
num=store.num_of_store()
candidate_index=[]
k_mean=[]
dist=[]
#随机产生k个中心点
for i in range(k):
    random_index=np.random.randint(0,num,1)
    if random_index in candidate_index:
        while random_index in candidate_index:
            random_index=np.random.randint(0,num,1)
        candidate_index.append(random_index.tolist()[0])
    else:
        candidate_index.append(random_index.tolist()[0])
    k_mean.append(store.list[0][random_index])

#把随机产生的中心点放入对应的集合中去
for i in range(k):
    store.list[0][candidate_index[i]]._set=i+1
    store.list[i+1].append(store.list[0][candidate_index[i]]._vector)

#把输入的点分配给随机产生的中心点
for i in range(num):
    if i in candidate_index:
        continue
    dist=[]
    for j in candidate_index:
        dist.append(((np.array(store.list[0][i]._vector) - np.array(store.list[0][j]._vector))**2).sum())
    min_dist_index=dist.index(min(dist))+1    #min_dist_index表示当前点里哪一个中心点最近
    store.list[0][i]._set=min_dist_index
    store.list[min_dist_index].append(store.list[0][i]._vector)    #把当前点加入到第k个集合中

#对每个类计算中心点，调整类中数据点位置，直到不再发生改变
flag=1  #flag=0表示数据点位置不再发生改变
iter_time=0
while flag==1:
    iter_time+=1
    flag=0  #初始状态为0
    candidate_mean_point = []
    for i in range(k):
        candidate_mean_point.append(np.array(store.list[i+1]).mean(0))
    for data in store.list[0]:
        dist=[]
        for j in candidate_mean_point:
            dist.append(((np.array(data._vector)-j)**2).sum())
        prev_index=data._set
        cur_index=dist.index(min(dist))+1
        if prev_index!=cur_index:       #有点进行位置调整
            flag=1
            data._set=cur_index         #将当前要调整位置的点的set属性设置为新的集合序号
            store.list[cur_index].append(data._vector)         #将当前调整的点放入到新的类中
            store.list[prev_index].remove(data._vector)        #删除原先类中的该点的数据

for i in range(len(store.list) - 1):
    print '第%s类为：' % (str(i + 1)), store.list[i + 1]

print 'iter_time:', iter_time
time2 = time.time()
print 'Running time:', time2 - time1


#如果聚类数大于8，则要增加下面两种点的标记类型
style=['or','ob','og','ok','oc','om','oy','ow']    #聚类点的标记类型
mean_mark=['Dr','Db','Dg','Dk','Dc','Dm','Dy','Dw']     #聚类中心的标记类型

#把不同类的点用不同颜色画出
for j in range(k):
    for point in store.list[j+1]:
        plt.plot(point[0],point[1],style[j])
#画出聚类中心
for j in range(k):
    plt.plot(candidate_mean_point[j][0],candidate_mean_point[j][1],mean_mark[j],markersize=10)
plt.title('kmean algorithm')
plt.show()