聚类分析算法

概念、原理： 参考文章一

基本的聚类分析算法

1. K均值：
基于原型的、划分的距离技术，它试图发现用户指定个数(K)的簇。

2. 凝聚的层次距离：
思想是开始时，每个点都作为一个单点簇，然后，重复的合并两个最靠近的簇，直到尝试单个、包含所有点的簇。

3. DBSCAN:
一种基于密度的划分距离的算法，簇的个数有算法自动的确定，低密度中的点被视为噪声而忽略，因此其不产生完全聚类。

一般数学建模使用K-Means 为主，所以下面主要介绍与K-means相关的所有代码。

图像

可视化聚类结果

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


def func01():    #生成二维随机点
    s = "8.60879466	214.2221436	34.31755286	5.913979217	7.681145748	4.019486375	10.90871211	11.70469991	20.05087612	0.064853127	273.350093	3.9550557	11.84476817	3.889739943	8.613913126	8.480019571	6.043974425	8.570848976	12.81918013	5.615367493	257.4091048	20.45940659	18.31377238	11.87079592	11.39314005	16.33021019	12.14806429	11.88364301	7.672279477	16.34202719	14.153227	8.415724972	7.681145748	14.84427145	11.70469991	7.681145748	203.2599058	11.14144184	12.43422417	7.366393142	3.841938626	5.565517548	19.19667605	19.57195838	16.23229793	3.930162208	15.52086972	7.681145748	14.96074993	7.681145748	3.868570788	5.607172981	7.681145747	7.343048335	11.3830478	8.367672595	269.2470718	11.25689469	4.054313049	59.77037809	17.90011733	152.9354127	8.895045192	11.47387804	217.9885334	16.41655679	10.99954235	10.99197317	49.95927129	5.097064639	282.4272835	191.8036898	169.0358292	1.215681537	203.0416609	3.992663098	11.24175632	7.681145748	8.598414461	11.03234216	9.507920341	7.681145748	11.10803402	70.05616308	19.18808516	-124.4069996	10.16290178	3.963490547	47.03815871	65.20500955	11.91195464	19.85413944	3.854230393	201.288385	14.11324229	252.10022	7.663750776	5.554591532	26.64990702	52.99999999"
    s1= s.split()
    price = []
    for i in s1:
        price.append(float(i))
    a = "8.59915836	212.5016154	34.3234387	6.041890605	7.681145748	4.000425951	16.25347365	11.70469991	20.4796643	0.065083235	272.5459715	3.909862801	13.14991884	3.851513389	8.613972305	8.452592827	6.180914799	8.56082937	12.80176732	4.592429361	258.5398214	20.17314157	12.02624826	11.70469991	11.38255442	16.72626344	12.11399426	11.70469991	7.681145748	15.93591562	13.56496877	8.414105233	7.681145748	14.72315408	11.70469991	7.681145748	203.526058	12.01488725	12.49618346	7.681145748	5.253114892	5.615367493	20.4796643	20.4796643	20.4796643	3.975389419	14.17620289	7.681145748	14.31319047	7.681145748	3.850297776	5.582545948	7.681145746	7.339833488	10.90871211	8.381382717	268.5445777	11.93948755	3.854552421	60.03251003	17.46615064	148.7804291	8.910644828	11.50157642	208.4861314	14.79152417	11.13777332	10.96036317	50.0033217	4.893293518	279.0054234	190.9863042	167.1789847	1.149011072	202.2500523	4.138530912	11.15798461	7.681145748	8.591217613	10.90871211	9.904016462	7.681145748	11.90355638	70.07694559	19.97210214	-124.4300188	10.29039835	3.9669966	47.34180303	227.1245276	13.06061354	20.68224836	3.967604407	201.4954977	16.02890384	260.5969943	7.646654963	5.615367493	27.4540285	52.99999998"
    a1 = a.split()
    increase = []
    for i in a1:
        increase.append(float(i))
    kjz1=[]
    for i in price:
        for j in increase:
            a=[i,j]
            kjz1.append(a)
    return kjz1


def func02(kjz2w): #绘图

    if kjz2w!=[]:
        colors=['b','g','r','c','m','y','k'];s=0;
        for j in kjz2w:
            for i in j:
                plt.plot(i[0], i[1], color=colors[s%len(colors)], marker='.')
            s=s+1;
        plt.ion();plt.show();plt.savefig('test1.jpg')#;plt.pause(5);plt.close();

def func03(kjz1,k):    #计算初始均值，并返回初始分组

    minxy=np.min(kjz1,axis=0).tolist();maxxy=np.max(kjz1,axis=0).tolist();
    xjg=(maxxy[0]-minxy[0])/k;yjg=(maxxy[1]-minxy[1])/k;
    meanxy=[];meanxy.append([minxy[0],minxy[1]]);meanxy.append([maxxy[0],maxxy[1]]);
    for j in range(1,k-1):
        meanxy.append([minxy[0]+xjg*j,minxy[1]+yjg*j])
    kjz2wxy2=[[] for j in range(0,len(meanxy))];
    for j in kjz1:
        s=0;lslb=[];
        for k in meanxy:
            lslb.append([s,(j[0]-k[0])**2+(j[1]-k[1])**2]);s=s+1; #一个坐标一组
        lslb.sort(reverse=False,key=operator.itemgetter(1))  #正序
        kjz2wxy2[lslb[0][0]].append(j)
    return kjz2wxy2

def func05(lb2): #剔除空列表

    j=0;
    while(True):
        if len(lb2[j])<=0:
            lb2.pop(j)
        else:
            j=j+1;
        if j>=len(lb2):
            break
    return lb2

def func06(kjz2wxy): #求组合中心(均值)

    meanxy=[];
    for j in kjz2wxy:
        meanxy.append(np.mean(j,axis=0).tolist())
    kjz2wxy2=[[] for j in range(0,len(meanxy))];
    for j in kjz2wxy:
        for i in j: #点
            s=0;lslb=[];
            for k in meanxy:
                lslb.append([s,(i[0]-k[0])**2+(i[1]-k[1])**2]);s=s+1;
            lslb.sort(reverse=False,key=operator.itemgetter(1))  #正序
            kjz2wxy2[lslb[0][0]].append(i)
    kjz2wxy2=func05(kjz2wxy2)
    return kjz2wxy2,meanxy

def func07(kjz2w,fz):

    kjz2wxy=func03(kjz2w,fz) #坐标列表，分组，0-按照x轴均分
    j=0;
    while(True):
        kjz2wxy,meanxy=func06(kjz2wxy)
        if j>0 and meanxy==meanxy2:
            break
        meanxy2=meanxy.copy();
        j=j+1;
    print('迭代%d次' % (j))
    func02(kjz2wxy) #绘图

if __name__=='__main__':

    start=time.time();
    for j in range(0,10):
        kjz2w=func01()
        func07(kjz2w,6) #分6组

    print('Time used:',int((time.time()-start)/60*10)/10,'分钟')

我们的 结果跑出来图像不是那么美丽，可能是题目数据不太ok。

聚类数目

然后又会遇到一个问题聚类的数目是多少呢？到底怎样比较好？
参考文章： 知乎聚类数目的确定

原理文章参考1
原理文章参考2
代码参考

可以选择这个图

还不知道怎么建立数据集_(:з」∠)_

代码

from sklearn.cluster import KMeans
from yellowbrick.cluster.elbow import kelbow_visualizer
from yellowbrick.datasets.loaders import load_nfl

X, y = load_nfl()

# Use the quick method and immediately show the figure
kelbow_visualizer(KMeans(random_state=4), X, k=(2,10))

或者这个图

在这里插入代码片

记得说明为啥选这个点哦~

聚类具体的分类情况

spssau自动处理： 在线spss