matlab下kmeans及pam算法对球型数据分类练习

clear all;
clc;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%数据初始化
Data=zeros(,);
%加噪声
for i=:
    Data(,i)=;
    Data(,i)=;
    Data(,i)=;
end
for i=:
    p=unifrnd(,);
    a=unifrnd(,*pi);
    b=unifrnd(,pi);
    Data(,i)=p*sin(a)*cos(b);
    Data(,i)=p*sin(a)*sin(b);
    Data(,i)=p*cos(a);
end
for i=:
    p=unifrnd(,);
    a=unifrnd(,*pi);
    b=unifrnd(,pi);
    Data(,i)=p*sin(a)*cos(b);
    Data(,i)=p*sin(a)*sin(b);
    Data(,i)=p*cos(a);
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%样本数量
[d,N]=size(Data);
%聚类的数目
K=;
%方法选择
method='kmeans';
%method='kmedoids';
%选取初始点
%max_Initial=max(,N/(*K));
max_Initial=;

label=zeros(max_Initial,N);
center=zeros(d,K,max_Initial);
C=zeros(,N);

%主循环
for initial_Case=:max_Initial
    pointK=Initial_center(Data,K);
    iter=;
    max_iter=1e+;
    % xK = pointK;
    disp(['------------KM进行第 ' num2str(initial_Case) ' 次重新选择初始中心-----------'])
    %%每次初始化K个中心点后，进行的循环
    while iter < max_iter
        iter = iter+;
        if mod(iter,)==
            disp([' 内部循环进行第 ' num2str(iter) ' 次迭代'])
        end
        %%%根据数据矩阵P中每个点到中心点的距离(最小)确定所属分类
        for i=:N
            dert = repmat(Data(:,i),,K)-pointK;
            distK=sqrt(diag(dert'*dert));
            [~,j] = min(distK);
            C(i) = j;
        end
        %%%重新计算K个中心点
        xK_=zeros(d,K);
        for i=:K
            Pi=Data(:,find(C==i));
            Nk = size(Pi,);
            % K-Means K-Medoids唯一不同的地方：选择中心点的方式
            switch lower(method)
                case 'kmeans'
                    xK_(:,i) = sum(Pi,)/Nk;
                case 'kmedoids'
                    Dx2 = zeros(,Nk);
                    for t=:Nk
                        dx=Pi-Pi(:,t)*ones(,Nk);
                        Dx2(t)=sum(sqrt(sum(dx.*dx,)),);
                    end
                    [~,min_ind] = min(Dx2);
                    xK_(:,i) = Pi(:,min_ind);
                otherwise
                    errordlg('请输入正确的方法：kmeans-OR-kmedoids','MATLAB error');
            end
        end
        %判断是否达到结束条件
        if xK_==pointK % & iter>
            disp(['###迭代 ' num2str(iter) ' 次得到收敛的解'])
            label(initial_Case,:) = C;
            center(:,:,initial_Case) = xK_;
            % plot_Graph(C);
            break;
        end
        pointK=xK_;
        %xK = xK_;
    end
    if iter == max_iter
        disp('###达到内部最大迭代次数1000,未得到收敛的解');
        label(initial_Case,:) = C;
        center(:,:,initial_Case) = xK_;
        %plot_Graph(C);
        %break
    end
end

%%%%增加对聚类结果最优性的比较
%距离差
dist_N = zeros(max_Initial,K);
for initial_Case=:max_Initial
    for k=:K
        tem=find(label(initial_Case,:)==k);
        dx=Data(:,tem)-center(:,k,initial_Case)*ones(,size(tem,));
        dxk=sqrt(sum(dx.*dx,));
        dist_N(initial_Case,k)=sum(dxk);
        %dist_N(initial_Case,k)=dxk;
    end
end

%%%%对于max_Initial次初始化中心点得到的分类错误
%%%%取错误最小的情况的Label作为最终分类
%求K类总的误差
dist_N_sum=sum(dist_N,);
[distmin,best_ind]=min(dist_N_sum);
%最佳分组
best_Label=label(best_ind,:);
%最佳中心
best_Center=center(:,:,best_ind);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%三维散布图
figure();
scatter3(Data(,:),Data(,:),Data(,:),'filled','cdata',best_Label);
title('Data Distribution');

function center=Initial_center(X,K)
%选取初始中心
N=size(X,);
rnd_Idx = randperm(N);
center = X(:,rnd_Idx(:K));
end