K-means聚类

Posted 2020-11-10 zuiaimiusi

 

1.随机产生k个分类特征的中心点

2.计算数据点到中心点的距离

3.数据点到哪个中心点最近就分到哪个类

4.迭代：更新中心点位置，重新计算距离并分配类别，直到总体距离最小

 

load fisheriris
figure;
speciesNum=grp2idx(species);
gscatter(meas(:,3),meas(:,4),speciesNum,[\'r\',\'g\',\'b\']);
xlabel(\'花瓣长度\');
ylabel(\'花瓣宽度\');
title(\'真实标记\');
set(gca,\'fontsize\',12);
set(gca,\'fontweight\',\'bold\');

data=[meas(:,3),meas(:,4)];
K=3;
%5此重复的全局最优解
[idx,cen]=kmeans(data,K,\'Distance\',\'sqeuclidean\',\'Replicates\',5,\'Display\',\'Final\');
%调整标号
dist=sum(cen.^2,2);
[dump,sortind]=sort(dist,\'ascend\');
newidx=zeros(size(idx));
for i=1:K
    newidx(idx==i)=find(sortind==i);
end
%花瓣长度和花瓣宽度散点图（kmeans分类）
figure;
gscatter(data(:,1),data(:,2),newidx,[\'r\',\'g\',\'b\']);
hold on
scatter(cen(:,1),cen(:,2),300,\'m\');
hold off
xlabel(\'花瓣长度\');
ylabel(\'花瓣宽度\');
title(\'kmeans分类\');
set(gca,\'fontsize\',12);
set(gca,\'fontweight\',\'bold\');