using Neural Network without toolbox

6 次查看（过去 30 天）

Mohammad 2022-12-2

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1869812-using-neural-network-without-toolbox

评论： Mohammad 2022-12-5

I have to write a code to model Neural Network. I write it with sigmoid function, back propogation, and gradient descent method.

My problem is that I can not insert input higher than 1.

This is my code:

X = (0:0.01:1.5);
X = X';
LX = length(X);
B_size = 1;
NO_B =  (LX / B_size);
Y_d = X.^2;
Width = 20;
H = zeros (Width,1);
H_f = zeros (Width,1);
Y = zeros(LX,1);
Y_f = zeros(LX,1);
W1 = rand (Width,B_size);
W2 = rand (B_size,Width); 
b1 = 1 ;
b2 = 1 ;
E_total = 1;
Eta = 0.1;
itt = 0;
epoch = 1500;
for e = 1 : epoch
    for i = 1 : NO_B
        itt = itt + 1; 
        XX = X(  (B_size * (i-1)) +1 : (i*B_size) );
        YY_d = Y_d(  (B_size * (i-1)) +1 : (i*B_size) );
        H = W1*XX + b1; 
        H_f = SIG(H);  
        Y = W2*H_f + b2;   
        Y_f = SIG(Y);  
        E_total = sum ( 0.5 * (( YY_d - Y_f ).^2)) ;
        E(itt) = E_total;
        ITT(itt) = itt;  
        delta =  YY_d - Y_f ;
        dY =  Y_f.*(1-Y_f) ;
        dH =  H_f.*(1-H_f) ;   
        pd2 = (delta.*dY) * H_f' ;   
        pd1 = (XX *((delta.*dY)' * W2).* dH')' ;    
        W2 = W2 + Eta*pd2;
        W1 = W1 + Eta*pd1;
        YY_f (  (B_size * (i-1)) +1 : (i*B_size) )= Y_f;   
    end
end
plot(X,YY_f,'r*',X,Y_d,'b:','LineWidth',2);
function [alpha_f] = SIG(alpha)
%SIGMOID FUNCTION
alpha_f = 1 ./ (1 + ((exp(1)) .^ (-alpha)));
end