//// Two dimensional self-organizing feature map al la Kohonen
 clear; nn=10; lambda=0.2; sig=2; sig2=1/(2*sig^2); 
 [X,Y]=meshgrid(1:nn,1:nn); 

 // Initial centres of prefered features:
 c1=0.5-.1*(2*rand(nn,nn)-1);
 c2=0.5-.1*(2*rand(nn,nn)-1);

//// training session
 for ntrial=1:1000 
    if(modulo(ntrial,100)==0) // Plot grid of feature centres
        clf; a=get("current_axes"); a.data_bounds=[0,0;1,1];
        plot(c1,c2,'k'); plot(c1',c2','k'); 
    end
    r_in=[rand();rand()];
    r=exp(-(c1-r_in(1)).^2-(c2-r_in(2)).^2);
    [rmax,x_winner]=max(max(r,'r')); [rmax,y_winner]=max(max(r','r'));
    r=exp(-((X-x_winner).^2+(Y-y_winner).^2)*sig2);
    c1=c1+lambda*r.*(r_in(1)-c1);
    c2=c2+lambda*r.*(r_in(2)-c2);
 end