how to modify the architecture of the network in matlab to build a Modular neural network

2 views (last 30 days)

dennis on 9 Sep 2014

0
Link

Direct link to this question

https://www.mathworks.com/matlabcentral/answers/154083-how-to-modify-the-architecture-of-the-network-in-matlab-to-build-a-modular-neural-network

Commented: Greg Heath on 9 Sep 2014

Hi my friend.

I'm trying to create a modular network; I train first several individual networks, but the last network need add a vector of output and I'm not how to do that. In Matlab ann are predefined network architectures.

My code is

%long is the longitude trained %regional is a vector data %sizemodule is the number of modules for module=1:1:sizemodule;

                  input(1:long)=regional(1:long,sizemodule);input=input';                
                  s11(1:long)=estp(1:long); 
                  target(1:long,1)= [s11'];                                           
                     input=con2seq(input');target=con2seq(target');
                                capa=[10 5];
                                          net = narnet(1:2,capa);
                                                %net = distdelaynet({1:4 1:2},capa);
                                                       %net.trainFcn = 'trainbr';  
                                                               net.trainFcn = 'trainlm';
                                                                      net = initlay(net)
                                                                            net.divideFcn = '';
                                                                      %net.biasConnect = [0;0];
                                                               net.trainParam.epochs = 1000;
                                                         net.trainParam.max_fail=100;
                                                  net.layers{1}.transferFcn = 'logsig';     %net.layers{2}.transferFcn = 'logsig';
                                            net.trainParam.goal=.00001;            %Performance goal
                                  %  net.trainParam.min_grad=0;           %Minimum performance gradient
                                  %net.trainParam.mu_max=10e100;     net.trainParam.max_fail=800;
                           net = train(net,input,target);
                           simulation(:,columna)=cell2mat(sim(net,input))';