Neural Network with learngdm as adaptation learning function

Hi all,

I need to apply a ANN to solve a regression problem. The network must have three inputs and one output. The network type is Feed-forward back propagation, the training function is trainlm, the adaptation learning function is learngdm, the performance function is MSE and the network could have one hidden layer. So far, I applied nntool to solve this problem but I later I have to use a for loop to built the network many times with different ANN parameters (e.g. number of hidden layers). Could anyone give me a small example of matlab code to construct a ANN with the mentioned characteristics? I am having problems to implement learndgm as learning function. A portion of the data I am using is attached to this post.

Thanks in advance.

Answer 1

[aditi bagora]

Hello Diego,

I understand that you are trying to build an ANN with hidden layer of varying sizes and following parameters.

• Training function: trainlm
• Learning function: learngdm
• Performance function: MSE

I am attaching a sample code that creates an ANN with the given parameters. You can use the “createnetwork()” function to define network with varying hidden layer sizes.

net = createnetwork([10,8]); % create a network
% number_hidden_layers is a vector of number of neurons required in each layer [10,8] means a network with 2 layers of sizes 10 and 8 respectively.
function net = createnetwork(number_hidden_layers)
   
   net = feedforwardnet(number_hidden_layers); % Specify the number of hidden layers and neurons per layer
  
   % Set the training function, adaptation learning function, and performance function
   net.trainFcn = 'trainlm';
   net.adaptFcn = 'learngdm'; 
   net.performFcn = 'mse';
   
   % Divide the data into training, validation, and testing sets
   net.divideFcn = 'dividerand';
   net.divideMode = 'sample'; 
   net.divideParam.trainRatio = 0.7; 
   net.divideParam.valRatio = 0.15;
   net.divideParam.testRatio = 0.15;
end

Hope this helps!

Regards,

Aditi