Python实现C-均值聚类

*不保证正确性，仅作记录

# -*- coding: utf-8 -*-
#############################
# Author : 水娃             #
# Date : 2020-12-04-01.21.51#
#############################
#导入库
import numpy as np
import matplotlib.pyplot as plt

#计算两个样本之间的距离
def calculateDistance(posa,posb):
    return np.sqrt(np.sum(np.power(posa-posb,2)))

#随机生成P个二维坐标
def getRandomData(P):
    return np.array([[np.random.rand()*100 for j in range(0,2)] for i in range(0,P)])

#获取初始类心
def getInitCenters(data,mp,P,C):
    #先把刚开始距离最远的两个点加入类心
    maxdis=0.0
    st1=-1
    st2=-1
    for i in range(0,P):
        for j in range(0,P):
            if mp[i][j]>maxdis:
                st1=i
                st2=j
                maxdis=mp[i][j]
    currentCenters=[st1,st2]
    
    #算出其余类心
    for i in range(0,C-2):
        currentCandidate=-1
        currentDistance=0;
        for j in range (0,P):
            if j in currentCenters:
                continue
            flag=True
            maxDistance=200
            for k in currentCenters:
                if mp[j][k]<currentDistance:
                    flag=False
                    break
                maxDistance=min(maxDistance,mp[j][k])
            if flag==False:
                continue
            if maxDistance>currentDistance:
                currentCandidate=j
                currentDistance=maxDistance
        currentCenters.append(currentCandidate)
    return currentCenters

#获取初始聚类结果
def getOriginalCluster(data,centers,mp,P,C):
    #按照最近距离原则将所有点聚类
    result=[[] for i in range(0,C)]
    for i in range(0,P):
        distance=mp[i][centers[0]]
        index=0
        for j in range(0,C):
            currentCenter=centers[j]
            if mp[i][currentCenter]<distance:
                distance=mp[i][currentCenter]
                index=j
        result[index].append(i)
    return result

#C均值聚类
def Cmeans(data,P,C):
    #获取距离列表，初始类心和初始聚类结果
    mp=[[calculateDistance(data[i],data[j])for j in range(0,P)]for i in range(0,P)]
    centers=getInitCenters(data,mp,P,C)
    cluster=getOriginalCluster(data,centers,mp,P,C)
    
    #将类心从索引转换成坐标
    positionCenters=[[]for i in range(0,C)]
    for i in range(0,C):
        positionCenters[i].append(data[centers[i]][0])
        positionCenters[i].append(data[centers[i]][1])
    centers=positionCenters
    
    #开始迭代聚类
    haschanged=True
    times=0
    while haschanged:
        times+=1
        print(times)
        tmpCluster=[[]for i in range(0,C)]
        haschanged=False
        
        #更新类心
        for i in range(0,C):
            totalx=0.0
            totaly=0.0
            for j in cluster[i]:
                totalx+=data[j][0]
                totaly+=data[j][1]
            centers[i][0]=totalx/(1.0*len(cluster[i]))
            centers[i][1]=totaly/(1.0*len(cluster[i]))
            
        #更新聚类结果
        for i in range(0,C):
            for j in cluster[i]:
                currentDistance=calculateDistance(centers[i],data[j])
                currentIndex=i;
                for k in range(0,C):
                    if calculateDistance(centers[k],data[j])<currentDistance:
                        currentDistance=calculateDistance(centers[k],data[j])
                        currentIndex=k
                if currentIndex!=i:
                    haschanged=True
                tmpCluster[currentIndex].append(j)
        cluster=tmpCluster
    return centers,cluster
    
#设定类的数目C和点数P
C=10
P=200

#获取随机数据
data=getRandomData(P)

#进行聚类获得结果
result,cluster=Cmeans(data,P,C)

#根据测试数据和结果点集绘制图形
for i in range(0,C):
    pointx=[]
    pointy=[]
    for k in cluster[i]:
        pointx.append(data[k][0])
        pointy.append(data[k][1])
    plt.scatter(pointx,pointy)
    
resultx=[result[i][0]for i in range(0,C)]
resulty=[result[i][1]for i in range(0,C)]
plt.scatter(resultx,resulty,marker='s',color='k')
plt.show()

本文地址：https://blog.csdn.net/qq_43807662/article/details/110633215