It is a prog about linear vector quantisation neural network. while updating weight when negative value comes as a result it is showing an error.we can make those negative values to zero and can proceed iteratrations. iam umable to do it.plz help me.

clc;
clear all;
st=[1 2 2 1 2];
alpha=0.6;
w=[0.2 0.8; 0.6 0.4; 0.5 0.7; 0.9 0.3];
disp('initial weight matrix');
disp(w);
x=[1 1 0 0; 0 0 0 1; 1 0 0 0; 0 0 1 1];
disp(x);
t=[st(2);st(3);st(4);st(5)];
e=1;
while (e<=3)
    i=1;
    j=1;
    k=1;
    disp('epoch=');
    e
    while(i<=4)
        for j=1:2
            temp=0;
            for k=1:4
                temp=temp+(w(k,j)-x(i,k))^2;
            end
            D(j)=temp;
        end
        if (D(1)<D(2))
            J=1;
        else
            J=2;
        end
        disp('The winning unit is');
        J
        disp('weight updation');
        if J==t(i)

              w(:,J)=w(:,J)+alpha*(x(i,:)'-w(:,J)); 
          else
              w(:,J)=w(:,J)-alpha*(x(i,:)'-w(:,J));

          end
          w
          i=i+1
      end
      temp=alpha(e);
      e=e+1;
      alpha(e)=0.5*temp;
      disp('first epoch completed');
      disp('learning rate updated for second epoch');
      alpha(e)
  end

[Walter Roberson]

Learn to use the debugger to figure out where the negative value is coming from.

After one loop iteration your alpha becomes a vector, and then your line

w(:,J)=w(:,J)+alpha*(x(i,:)'-w(:,J))

starts involving matrix multiplication where the "*" is. Are you sure that is what you want, not element-by-element multiplication, the .* operator ?

Answer 1

[Greg Heath]

The topic should be LEARNING (NOT linear) VECTOR QUANTIZATION

Why is st 5 dimensional?

alpha = 0.6 is too high for an initial learning rate

e = 3 is too low for a maximum number of epochs

Why are you using loops instead of vectorization?

Since the x and w vectors have the same dimensions, (w(k,j)-x(i,k))^2 is incorrect

Your treatment of alpha as a scalar and a vector is incorrect.

HTH

Thank you for formally accepting my answer

Greg