User define function missbehave...

Question

Shivang Patel 2015-3-12

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/182756-user-define-function-missbehave

编辑： Shivang Patel 2015-3-12

Hi... I m working with BPNN, And as a reference code i m using this... " ANOOP ACADEMIA – BLOG's BPNN Code "

My problem is, in the for loop after function execute... value's are not updated and result are same... Where is the problem with it... Everything looks fine... I m confuse...where is my mistake... Code is...below

% Assigning the number of hidden neurons in hidden layer
m = 2;
errorValue_theshold = 100;
errorValue = errorValue_theshold + 1; % Only for initial...
delta_V = 0;
delta_W = 0;
[l,b] = size(data);
[n,a] = size(target);
V = rand(l,m); % Weight matrix from Input to Hidden
W = rand(m,n); % Weight matrix from Hidden to Output
count = 0;
itration = 100;
for count = 1: itration
      [errorValue delta_V delta_W] = trainNeuralNet(data,target,V,W,delta_V,delta_W);
      count = count + 1;
      fprintf('Error : %f\n', errorValue);
      fprintf('#itration : %d \t', count);
      Error_Mat(count)=errorValue;
      W = W + delta_W; 
      V = V + delta_V;    
      if errorValue < errorValue_theshold
          fprintf('\n\nFinal Error : %f\n', errorValue);
          fprintf('#totalItration : %d\n', count);
          Error_Mat(count)=errorValue; 
      end
  end

Function Code is : ----------------

          *function [errorValue delta_V delta_W] = trainNeuralNet(Input, Output, V, W, delta_V, delta_W)*             
            Output_of_InputLayer = Input;            
            Input_of_HiddenLayer = V' * Output_of_InputLayer;  % % netj             
            [m n] = size(Input_of_HiddenLayer);
            Output_of_HiddenLayer = 1./(1+exp(-Input_of_HiddenLayer));
            Input_of_OutputLayer = W'*Output_of_HiddenLayer;
            clear m n;             
            [m n] = size(Input_of_OutputLayer);
            Output_of_OutputLayer = 1./(1+exp(-Input_of_OutputLayer));
            difference = Output - Output_of_OutputLayer; 
            square = difference.*difference; 
            errorValue = sqrt(sum(square(:)));             
            clear m n 
            [n a] = size(Output);
            for i = 1 : n 
                for j = 1 : a
                    d(i,j) =(Output(i,j)-Output_of_OutputLayer(i,j))*Output_of_OutputLayer(i,j)*(1-Output_of_OutputLayer(i,j));
                end  
            end
            Y = Output_of_HiddenLayer * d'; 
            if nargin == 4
                delta_W=zeros(size(W)); 
                delta_V=zeros(size(V));
            end
            etta=0.6; alpha=1;
            delta_W= alpha.*delta_W + etta.*Y;             
            error = W*d;
            clear m n 
            [m n] = size(error);
            for i = 1 : m
                for j = 1 :n
                    d_star(i,j)= error(i,j)*Output_of_HiddenLayer(i,j)*(1-Output_of_HiddenLayer(i,j));
                end
            end
            X = Input * d_star';
            delta_V = alpha * delta_V + etta * X;
            end