I want to define matlab artificial neural network initial weights to initialize with Genetic Algorithm. How can i change train function for this?

I want to define matlab artificial neural network initial weights to initialize with Genetic Algorithm. How can i change train function for this?

Answer 1

[Akshat]

Hi @mahmud sadik,

I am assuming you have a fitness function defined to run the Genetic Algorithm. In my opinion, you do not need to modify the train function, you can just set the weight and biases before training the network.

Here is some code to assist you further:

% Define a simple feedforward neural network
hiddenLayerSize = 10;
net = feedforwardnet(hiddenLayerSize);

numWeightsBiases = sum([net.layers{:}.size]) * net.input.size + ...
                   sum([net.layers{:}.size].^2) + ...
                   sum([net.layers{2:end}.size]);

function error = fitnessFcn(weightsBiases, net, X, T)
    net = setwb(net, weightsBiases);
    Y = net(X);
    error = perform(net, T, Y);
end

X = ...; % Input features
T = ...; % Target labels
options = optimoptions('ga', 'Display', 'iter', 'MaxGenerations', 100);

[optWeightsBiases, fval] = ga(@(wb) fitnessFcn(wb, net, X, T), numWeightsBiases, [], [], [], [], [], [], [], options);

net = setwb(net, optWeightsBiases);
% Train the network further (optional)
net = train(net, X, T);
Y = net(X);

Feel free to follow-up in case you have any further queries!