clusang.m

最新的模式识别分类工具箱,希望对朋友们有用!

% ---------------------------------------------------------------- 
% FUNCTION   clusang.m    CLUSterin algorithm, ANGle metric.
% ---------------------------------------------------------------- 
% 		builds a block training set. 
% Usage: 
%   cluster_centers = clusang(orig_cluster_centers, training set, precision);
% 
% orig_cluster: matrix, each ROW is a cluster center.
% training set: matrix, each ROW is a sample.
% Precision = stopping criterion: If the maximum distance between the centers
%             of the clusters is less than precision, then stop;
%
% __________________________
%  Author: Vittorio Castelli
%  Copyright IBM T. J. Watson Research Center
%      January 27, 1995;	Last modified: January 27, 1995

function clCen =  clusang(origCen, trSet,prec);

learning_rate = 1;

[m,n] = size(trSet);
[k,n] = size(origCen);
aaa = ones(1,n);
dist = 1;
n_dist = 32;

clCen = origCen;
count = 0;
distance = zeros(1,n_dist);
while dist > prec,
  oldCen=clCen;
  dist = trSet * clCen';
  dist = dist';
  [dist,ord] = sort(dist);
  ord = ord';

  ord = ord(:,k);
  dist=0;
  for i=1:k,
        if(sum(ord==i)==0),
             fprintf(' %g',i)
        end
         m = mean((((ord==i)*aaa).*trSet));
         clCen(i,:) = normalize(clCen(i,:)+m.*learning_rate);
         d_i = clCen(i,:)*oldCen(i,:)';
         d_i = acos(d_i);
         dist=max(dist, d_i);
   end
   
   count = count+1;
   distance(count)= dist;
   if rem(count,n_dist)==0,
      if ceil(count/n_dist/2)-floor(count/n_dist/2) ==0,
          figure(2);
      else
          figure(1);
      end
      plot(distance,'+');
      grid
   end

end
fprintf('\n');