function [C,V,L]=optkmeansr(pattern,centers,var,learn_rate,alpha)
%OPTKMEANSR Recursive version of the optimal adaptive k-means clustering algorithm
%
%  [C,V,L]=OPTKMEANSR(pattern,centers,var,learn_rate,alpha)
%
%    C - New centers computed
%    V - Variations in each cluster
%    L - New learning rate computed
%
%    pattern - Input pattern present in one iteration
%    centers - Centers computed in previous iteration
%    var - Cluster variations in previous iteration
%    learn_rate - Learning rate in previous iteration
%    alpha - Parameter (very close to 1.0 ; default is 0.9999)
%
%  Computes new values for the centers and cluster variation, associated with
%  each cluster, given the previous iteration values and a new input pattern.
%
%  References:
%
%    Chinrungrueng, C., and C.H. Séquin (1995).
%    "Optimal Adaptive k-means Algorithm With Dynamic Adjustment of Learning Rate"
%    IEEE Transactions on Neural Networks, vol.6, Nº 1, pp. 157-169
%
%  04/06/1999 ; Pedro Frazão Ferreira
%

if (nargin<4) | (nargin>5),
   error('Wrong number of input arguments. Try "help optkmeansr".');
end
if nargin==4,
   alpha=0.9999;
end

dim=size(centers);
C=centers;
V=var;
for i=1:dim(1),
   pattern(i,:)=pattern(1,:);
end
distances=sum(abs(pattern-centers).^2,2);
wdistances=var.*distances;
mship=find(wdistances==min(wdistances));
C(mship,:)=centers(mship,:)+(learn_rate*(pattern(mship,:)-centers(mship,:)));
V(mship)=(alpha*var(mship))+((1-alpha)*distances(mship));
vknorm=V/sum(V);
L=1-(sum(-vknorm.*log(vknorm))/log(dim(1)));