%  CmpInit.m
%        A.P. Paplinski   27 August 2004

% Initialisation of competitive learning
% Generation of a 2-D pattern consisting of  m  clusters
%   of normally distributed points

 clear

p = 2; m = 5 ;      % p inputs, m  outputs
clst = randn(p, m); %  cluster centroids
clst = [1 2 4 6 5 
        5 1 4 2 6] ;

Nk = 20;          % points per cluster
N = m*Nk ;        % total number of points

sprd = 0.5 ;      % a relative spread of the Gaussian "blob"
X = zeros(p,N+m); % X is  p by m+N input data matrix
wNk = ones(1, Nk);

for k = 1:m
  xc = clst(:,k) ;
  X(:,(1+(k-1)*Nk):(k*Nk)) = sprd*randn(p,Nk)+xc(:,wNk) ;
end

[xc k] = sort(rand(1,N+m));
X = X(:, k) ;      % input data is shuffled randomly
Win = X(:,1:m)';   % Initial values of weights
X = X(:,m+1:N+m);

% figure(1), clf reset
% plot(X(1, :), X(2, :), 'g.', Win(:, 1), Win(:, 2), 'b^', ...
%   clst(1, :), clst(2, :), 'ro' ), grid on, 
%   title('Afferent data'), xlabel('x_1'), ylabel('x_2')