mykcentres.m

支持向量域是近几年采用的一种较新的分类器

% KCENTRES Find k centres objects from distance matrix
%
%	[labels,J,dmin] = kcentres(D,k,n)
%
% If D is a square distance matrix between m objects then J is the set of centre
% points, i.e. the subset of k objects that minimizes dmin, the maximum of the
% distances over all objects to the nearest centre point. For k > 1 the results
% depend on a random initialisation. The procedure is repeated n times and the
% best result is returned. In labels a set of m object labels is stored in
% which for each object the nearest of the centre points is returned. 
%
%	J = kcentres(D)
%	J = kcentres(D,1)
%
% In case k = 1 just the centre point is returned.
%
% Default: k = 1, n = 1.
%
% See also soset, kmeans

% Copyright: R.P.W. Duin, [e-mail]
% Faculty of Applied Physics, Delft University of Technology
% P.O. Box 5046, 2600 GA Delft, The Netherlands


function [labels,Jopt,dopt] = mykcentres(d,k,n);
if nargin < 3, n = 1; end
if nargin < 2, k = 1; end
[m,m] = size(d);
if k == 1
	dmax = max(d);
	[dm,labels] = min(dmax);
	return
end
dopt = inf;
for try = 1:n
M = randperm(m); M = M(1:k);
J = zeros(1,k);
while 1
	[dm,I] = min(d(M,:));
	for i = 1:k
		JJ = find(I==i);
		j = kcentres(d(JJ,JJ));
    if size(J(i))~=size(JJ(j))
      JJ
      j
      J(i)
      JJ(j)
    end
    if ~isempty(j)
      J(i) = JJ(j);
    else
      J(i) = floor(m*rand)+1;
    end
	end
	if all(M == J)
		[dmin,labs] = min(d(J,:));
		dmin = max(dmin);
		break;
	end
	M = J;
end
if dmin < dopt
	dopt = dmin;
	labels = labs';
	Jopt = J;
end
end