小波神经网络MATLAB程序

分享一下我老师大神的人工智能教程！零基础，通俗易懂！http://blog.csdn.net/jiangjunshow

也欢迎大家转载本篇文章。分享知识，造福人民，实现我们中华民族伟大复兴！

               

clear all
%initiate of data
 P=3 %numberof sample
 m=1%number of input node
 n=10%number of hidden node
 N=1%number of ouptut node
 %
 %a(n) b(n) scale and shifting parameter matrix
 %x(P,m) input matrix of P sample
 %net(P,n) ouput of hidden node
 %y(P,N) output of network
 %d(P,N) ideal output of network
 % phi(P,n) ouput of hidden node wavelet funciton
 %W(N,n)weight value between ouput and hidden
 %WW(n,m) weight value between  hidden and input node
x=[4;5;6]
d=[1.3;3.6;6.7]
W=rand(N,n)
WW=rand(n,m)
a=ones(1,n)
for j=1:n
b(j)=j*P/n;
end
 %%%%%%%%%%%%%%%%%%
 %EW(N,n) gradient of W
 %EWW(n,m) gradient of WW
 %Ea(n) gradient of a
 %Eb(n) gradient of b
 %%%%%%%%%%%%%%]
 epoch=1;
 epo=100;
 error=0.05;
 err=0.01;
 delta =1;
 lin=0.5;
 while (error>=err & epoch<=epo)
    
     u=0;%u is the middle variant
     %caculation of net input
     for p=1:P
         for j=1:n
             u=0;
              for k=1:m
                  u=u+WW(j,k)*x(p,k);
              end
              net(p,j)=u;
             
         end
     end
     %calculation of morlet 0r mexican wavelet output
          for p=1:P
              for j=1:n
                  u=net(p,j);
                  u=(u-b(j))/a(j);
                  phi(p,j)=cos(1.75*u)*exp(-u*u/2); %morlet wavelet
                  %phi(p,j)=(1-u^2)*exp(-u*u/2); %mexican hat wavelet
              end
          end
      %calculation of output of network
    
         for p=1:P
             for i=1:N
                 u=0;
                 for j=1:n
                     u=u+W(i,j)*phi(p,j);
                 end
                 y(p,i)=delta*abs(u);
             end
         end
       %calculation of error of output
       u=0;
     for p=1:P
         for i=1:N
            u=u+abs(d(p,i)*log(y(p,i))+(1-d(p,i)*log(1-y(p,i))));
            %u=u+(d(p,i)-y(p,i))^2;
         end
     end
     %u=u/2
     error=u;
     %calculate of gradient of network
     for i=1:N
         for j=1:n
             u=0;
             for p=1:P
                 u=u+(d(p,i)-y(p,i))*phi(p,j);
             end
             EW(i,j)=u;
             %EW(i,j)=-u;%the resule would be wrong
         end
     end
     for j=1:n
         for k=1:m
             u=0
             for p=1:P
                 for i=1:N
                    u=u+(d(p,i)-y(p,i))*W(i,j)*phi(p,j)*x(p,k)/a(j) ;
                 end
             end
             EWW(j,k)=u;
             %EWW(j,k)=u the result would be wrong
         end
     end
     for j=1:n
             u=0
             for p=1:P
                 for i=1:N
                    u=u+(d(p,i)-y(p,i))*W(i,j)*phi(p,j)/a(j) ;
                 end
             end
             Eb(j)=u;
     end
     for j=1:n
             u=0
             for p=1:P
                 for i=1:N
                    u=u+(d(p,i)-y(p,i))*W(i,j)*phi(p,j)*((net(p,j)-b(j))/b(j))/a(j) ;
                 end
             end
             Ea(j)=u;
     end
    %adjust of weight value
    WW=WW-lin*EWW;
    W=W-lin*EW;
    a=a-lin*Ea;
    b=b-lin*Eb;
     %number of epoch increase by 1
     epoch=epoch+1;
    
 end
 

           

给我老师的人工智能教程打call！http://blog.csdn.net/jiangjunshow