backpropagation

[t.man]

hi,
i'm trying to implement backpropagation network to learn certain functions.
but my error stay above something like 200. where may i be doing wrong?

here is the matlab code. 2 inputs, 3 hidden nodes and 2 output nodes.

echo on
clc

clear all
x = [0 1 2 3 4 
    -2 -1 0 1 2];
d=[-1 3.736 3.271 3.100 9.037
    15.000 13.000 15.000 21.000 31.000];
 
% x = [0 1 2
%     -2 1 2];
% d=[-1 5 11.27
%    0 1 0];
%weights connecting inputs to the hidden node
 iw = zeros(3,2);
%weights connecting hidden nodes with the output nodes
 lw= zeros(2,3);

%learning rate
 mu = 0.25;

%test parttens
 max_tests = 5;

 echo off;pause
 mse = inf;
 a = 0;
 tic
 while(mse > 0.26)
     sum = 0;
    a = a+1;
     for n=1:max_tests,
         
        a_in=iw*x(:,n);
       %outputs vector for the hidden nodes
        a=[1/(1+exp(-a_in(1))) 1/(1+exp(-a_in(2))) 1/(1+exp(-a_in(3)))];
        
        %input to the output nodes
        y_in = lw*a';
        
        %output vector
        y = [1/(1+exp(-y_in(1)))  1/(1+exp(-y_in(2)))];
        
        %error vector
        e = d(:,n)'-y
        
        %theta for the outputs
        thetaout=[e(1)*y(1)*(1-y(2));
        e(2)*y(2)*(1-y(2))];
        
        %theta for the hidden nodes
        thetaL1 = a(1)*(1-a(1))*lw(:,1)'*thetaout;
        thetaL2 = a(2)*(1-a(2))*lw(:,2)'*thetaout;
        thetaL3 = a(3)*(1-a(3))*lw(:,3)'*thetaout;
        
        %updates the weights
        lw(1,:)=lw(1,:)+mu*e(1)*a;
        lw(2,:)=lw(2,:)+mu*e(2)*a;
        iw(1,:)=iw(1,:)+mu*thetaL1*x(:,n)';
        iw(2,:)=iw(2,:)+mu*thetaL2*x(:,n)';
        iw(3,:)=iw(3,:)+mu*thetaL3*x(:,n)';
                
        %mean square erro preparation
        sum= sum+(e(1)*e(1)+e(2)*e(2))/2;
        ['iteration #: ' int2str(a) ':']


%         pause
     end
    %calculate mean square error
    mse=sum/max_tests
 end
 toc
 iw
lw

 

[smanches]

You might try the Robotics topic, but this forum doesn't deal with the more advanced AI features much. An actual AI forum would give you a much better response.