clc
clear
close all
tic
%先形成初始点
center=[1,1,1,1,1;10,10,10,10,10;100,100,100,100,100;1000,1000,1000,1000,1000];
% center=[1;10;100;1000];
dataN=100;% 生成多少个数据
Dim=5;
clusterN=10;
% data=zeros(Dim,dataN);
% % data=[11,101,1001,1,2,3];
% for I=1:dataN
%     i=round(1+(4-1)*rand());
%     data(:,I)=center(i)*( -1+(1+1)*rand(Dim,1));
% end

data=rand(Dim,dataN);
data=load('data');
data=data.data;

SetS=[1;];

for clusterI=1:clusterN-1
    maxG=-100*ones(dataN,1);
    for cluster=1:dataN
        if sum(ismember(cluster,SetS))>0
            continue
        end
        %选一个候选数据
        cddtI=cluster;
        cadSetS=[SetS;cddtI;];
        Cij=0;
        for J=1:dataN
            if sum(ismember(J,cadSetS))>0
                continue
            end
            d=data(:,J);
            %寻找最短距离
            matDistance=repmat(d,1,length(SetS));
            minD=matDistance-data(:,SetS);
            minD=diag(minD'*minD);
            minD=min(minD.^.5);
            
%             distanceIJ=sum((d-data(:,cddtI)).^2).^.5;
            distanceIJ=metricFun( d,data(:,cddtI) );
            Cij=Cij+max([minD-distanceIJ,0]);
        end
        maxG(cluster)=Cij;
    end
    maxGInd=find(maxG==max(maxG));
    SetS=[SetS;maxGInd(1)];
end
SetS
%% 进入SWAP部分
% SetS=[1;2;3];
USetS=setxor(1:dataN,SetS);

while 1
    minKjih=1e20;
    for I=1:length(SetS)
        for H=1:length(USetS)
            %交换两个集合的元素
            cadSetS=SetS;
            cadUSetS=USetS;
            swap=cadSetS(I);
            cadSetS(I)=cadUSetS(H);
            cadUSetS(H)=swap;
            sumKjih=0;
            for D=1:length(cadUSetS)
                J=cadUSetS(D);
                d=data(:,J);
                if J==cadSetS(I)
                    continue
                end
                matDistance=repmat(d,1,length(SetS));
                minD=matDistance-data(:,SetS);
                minD=diag(minD'*minD).^.5;
                minD_t=min(minD);
                minD(minD==minD_t)=1e20;
                min2D=min(minD);
                minD=minD_t;            
%                 distanceIJ=sum((d-data(:,cadUSetS(H))).^2).^.5;%S(I)已经等于U(H)
                distanceIJ=metricFun( d,data(:,cadUSetS(H)) );
%                 distanceHJ=sum((d-data(:,cadSetS(I))).^2).^.5;
                distanceHJ=metricFun( d,data(:,cadSetS(I)) );
                if distanceIJ>minD
                    Kjih=min([distanceHJ-minD,0]);
                end
                if abs(distanceIJ-minD)<1e-5
                    Kjih=min([distanceHJ,min2D])-minD;
                elseif distanceIJ<minD
                    fprintf('Input must be a string\n')
                end

                sumKjih=sumKjih+Kjih;
            end
            if sumKjih<minKjih
                minKjih=sumKjih;
                minSetS=cadSetS;
                minUSetS=cadUSetS;
            end
        end
    end
    if minKjih<0
        minKjih
        SetS=minSetS;
        USetS=minUSetS;
        SetS
    else
        minKjih
        SetS
        fprintf('clustering is done.\n')
        break;
    end
end
toc