%%example76
%%training samples
in=10;
num=in;

b0(:,1)=reshape(double(rgb2gray(img1)),32*32,1);
b0(:,2)=reshape(double(rgb2gray(img2)),32*32,1);
b0(:,3)=reshape(double(rgb2gray(img3)),32*32,1);
b0(:,4)=reshape(double(rgb2gray(img4)),32*32,1);
b0(:,5)=reshape(double(rgb2gray(img5)),32*32,1);
b0(:,6)=reshape(double(rgb2gray(img6)),32*32,1);
b0(:,7)=reshape(double(rgb2gray(img7)),32*32,1);
b0(:,8)=reshape(double(rgb2gray(img8)),32*32,1);
b0(:,9)=reshape(double(rgb2gray(img9)),32*32,1);
b0(:,10)=reshape(double(rgb2gray(img10)),32*32,1);


%%initialize the parameters
cpts=8;
W=rand(cpts,32*32)*1e-3;
eta=str2num(rate);
maxsteps=str2num(epoch);

%from each column subtract the mean of the traing samples
b=sum(b0');
b=b'/10;
b0=b0-[b b b b b b b b b b ];

%%training the weight matrix with the GHA(Sanger) algorithm
for n=1:maxsteps     
    eta=eta;
    y=zeros(cpts,10);   
    for r=1:cpts
        y(r,:)=W(r,:)*b0 ;              
        b2=W'*y;        
        dW(r,:)=eta*y(r,:)*(b0'-b2');
        W(r,:)=W(r,:)+dW(r,:);
    end
    E(n)=norm(b0-b2);
    E0=E(n);
    
    if (E0<1) 
        break, end;
    step=n;    
end

x=1:maxsteps;
E=E(x);    
figure('name','網絡訓練過程圖示','numbertitle','off');
plot(x,E),title('Target Functioin');