leader_follower_matrix.m

leader-follower聚类算法的matlab实现

function [xy_matrix,w_class,w_class_num,class_number]=leader_follower_matrix(xy_matrix,i,w_class,w_class_num,constant)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%leader_follower algorithm: (process one sample each iteration)
%                         to  get the class whitch has nearest distance
%                         (smaller than the threhold) to the new sample, 
%                         and conflate the new sample to the very  class.
%                         if the minimum distance is higher than the
%                         threhold,the make the new sample a independent
%                         class.
%    input value:                  
%                          sub_matrix:     each row is a sample 
%                          w_class:    contain the cluster centers which have been found                        
%                          w_class_num:contain the number of each class which has been found 
%                          constant:   the threhold value defined before the program
%    output value:
%                          sub_matrix:       include the message of classID 
%                          w_class:      the updated cluster centers
%                          w_class_num:  the updated numbers of each class
%                          class_number: current number of whole classes
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
old_class_size=size(w_class,1);
sample_size=size(xy_matrix,1);
for class_id=1:old_class_size
    ww(1,class_id)=sqrt(double((w_class(class_id,1)-xy_matrix(i,1))^2+(w_class(class_id,2)-xy_matrix(i,2))^2));
end
new=min(ww);%当前样本聚类各个聚类中心的最小值
if (new<=constant)
num_id=0;
  for class_id=1:old_class_size
      if (new==ww(1,class_id))  % 找到距离样本点最近的聚类中心class_id及其个数num_id+1
          num(1,(num_id+1))=class_id;%将与样本距离相等的聚类中心的id保存在num中(1*n)
          num_id=num_id+1;           %num_id即将合并的聚类中心数目
      end
   end


   sum_value=zeros(1,2);
   sum_p=0;

% if (new<=constant)
 for id=1:num_id  %num()是变化前的类别标识符
         temp_1_matrix=num(1,id)*ones(sample_size,1);        
         L1=(xy_matrix(:,3)==temp_1_matrix);                 
         xy_matrix(:,3)=xy_matrix(:,3)+(-(num(1,id)+1))*L1;
%                                                                              
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%$$$$$$ end  %%%将即将被合并的样本之类别标记改为-1
     iden=max(xy_matrix(:,3));
     if num(1,id)<iden%%%%将要合并得标志符不是最大标志符时：
             change=num(1,id)+1;                                           
             temp_2_matrix=change*ones(sample_size,1); % matrix calculalte 
             L2=(xy_matrix(:,3)>=temp_2_matrix);                           
             xy_matrix(:,3)=xy_matrix(:,3)+(-1)*L2;                     
                                                                           
                                                                           
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     end  %%%将其他类的标识符减少1 
     p=w_class_num(num(1,id),1);                 %得到即将合并各类的已归类样本数目
     sum_value=sum_value+p*w_class(num(1,id),:); %计算即将合并各类的所有样本之和
     sum_p=sum_p+p;                              %合并的各类中样本数目之和
     w_class_num(num(1,id),:)=[];                %将被合并的样本数目去掉
     w_class(num(1,id),:)=[];                    %将被合并的样本的聚类中心去掉

 end
 w_new=double((sum_value+xy_matrix(i,1:2))/(sum_p+1));   %计算出新类别的聚类中心
%  w_class=[w_class;w_new];            %将新聚类中心加到聚类中心矩阵末尾
 w_class=vertcat(w_class,w_new); 
 w_class_num=[w_class_num;(sum_p+1)];%将新类别样本数加到各个类别样本数矩阵末尾
%  w_class_element

else %如果距离当前所有聚类中心距离都大于阈值，则自立一类
%     w_class=[w_class;sample];  
    w_class=vertcat(w_class,xy_matrix(i,1:2)); 
    w_class_num=[w_class_num;1];
end
    class_number=size(w_class,1);%得到所有类别标识总数&&也是当前得到的新类别的标识