Weights during network training

Dear users,

I am training a neural network using traingda algortithm. Is it possible to access the weights of the network at the intermediate epochs? With the commands net.IW, net.LW, and net.b I can access the weights only at the end of the training. I would like to access all the intermediate weights.

Any help is really appreciated! Thank you.

Mauro

Answer 1

[Greg Heath]

You will have to loop over an inner one-epoch loop.

Hope this helps.

Greg

Answer 2

[Mauro]

Dear Greg, thank you for your reply. I am not sure I have understood what you mean. Can you give me more details, please? How can I loop over an inner one-epoch loop?

Thank you,

Mauro

Answer 3

[Greg Heath]

rand('state',0);

net = newff(p,t,H);

IW = net.IW{1,1};

b1 = net.b{1};

IW = net.LW{2,1};

b2 = net.b{2};

% Store and/or plot the weight values

MSEgoal = mean(var(t'))/100

Nepochs = 1

Nloops = 100

net.trainParam.goal = MSEgoal;

net.trainParam.epochs = Nepochs;

net.trainParam.show = inf;

for i = 1:Nloops

    net = train(net,p,t);

    IW = net.IW{1,1};

    b1 = net.b{1};

    LW = net.LW{2,1};

    b2 = net.b{2};

    % Update storage and/or plots

end

REPRODUCIBILITY WARNING:

Using Nepochs = 100 with Nloops = 1 yields different results because each call of TRAIN reinitializes internal parameters.

Hope this helps.

Greg