神经网络Version2

% A demo to BP nerual networkformat long% define the sigmoid functionf = @(x) 1/(1+exp(-x));% enter the learning rateEta = 0.5;x=[0.05,0.10];y=[0.01,0.99];% modifiedb=[1,1,1];bw=[0.35,0.6,0.35];temp = size(x); % number of rowsm = temp(1);% number of colsn = temp(2);% the weightsw=[0.15,0.20,0.25,0.30;    0.40,0.45,0.50,0.55;    0.40,0.45,0.50,0.55];% gradient gradient = zeros(3,4);% the output of neurons in hidden layerhout = zeros(2,2);% the output of neurons in hidden layeroout=[1;1];% initialize the condition to terminate loopdiff = 1;while(diff <= 1)        % 正向推理（输入层->隐藏层1）        neth1=w(1,1)*x(1)+w(1,2)*x(2) + b(1)*bw(1);        hout(1,1) = f( neth1 );        hout(1,2) = f( w(1,3)*x(1)+w(1,4)*x(2) + b(1)*bw(1) );        %(隐藏层1->隐藏层2)        hout(2,1)=f( w(2,1)*hout(1,1)+w(2,2)*hout(1,2) + b(2)*bw(2) )        hout(2,2)=f( w(2,3)*hout(1,1)+w(2,4)*hout(1,2) + b(2)*bw(2) )        %(隐藏层2->输出层)        oout(1)=f(  w(3,1)*hout(2,1)+w(3,2)*hout(2,2) + b(3)*bw(3) );        oout(2)=f(  w(3,3)*hout(2,1)+w(3,4)*hout(2,2) + b(3)*bw(3) );        % gradient of output layer        deltaO=zeros(2,4);        deltaO(1,4)=( oout(1) - y(1) ) * oout(1) * (1 - oout(1) );        deltaO(2,4)=( oout(2) - y(2) ) * oout(2) * (1 - oout(2) );        %求输出层到隐藏层2的偏导        gradient(3,1)=deltaO(1,4)*hout(2,1);        gradient(3,2)=deltaO(1,4)*hout(2,2);        gradient(3,3)=deltaO(2,4)*hout(2,1);        gradient(3,4)=deltaO(2,4)*hout(2,2);        %求deltaO (3)        deltaO(1,3) = ( deltaO(1,4)*w(2,1) + deltaO(2,4)*w(2,3) )*hout(2,1)*(1-hout(2,1));        deltaO(2,3) = ( deltaO(2,4)*w(2,4) + deltaO(1,4)*w(2,2) )*hout(2,2)*(1-hout(2,2));        gradient(1,:);        %w(2,:)           %求隐藏层2到隐藏层1的偏导        gradient(2,1)= deltaO(1,3) * hout(1,1);        gradient(2,2)= deltaO(1,3) * hout(1,2);        gradient(2,3)= deltaO(2,3) * hout(1,1);        gradient(2,4)= deltaO(2,3) * hout(1,2);        %求deltaO (2)        deltaO(1,2) = ( deltaO(1,4)*w(2,1) + deltaO(2,4)*w(2,3) )*hout(1,1)*(1-hout(1,1));        deltaO(2,2) = ( deltaO(2,4)*w(2,4) + deltaO(1,4)*w(2,2) )*hout(1,2)*(1-hout(1,2));        %求隐藏层1到输入层的偏导        gradient(1,1)= deltaO(1,2) * x(1);        gradient(1,2)= deltaO(1,2) * x(2);        gradient(1,3)= deltaO(2,2) * x(1);        gradient(1,4)= deltaO(2,2) * x(2);%         gradient(1,:);        %进行梯度下降        w=w-Eta*gradient    % update diff    diff=diff+1;end