Hidden layer activations with Neural Network Toolbox

Hello,

I have recently started using Matlab's Neural Network Toolbox, after some years of working with Netlab. Does anybody know how to get the activations (=output of the transfer function) of the hidden layers?

So if I create a 4-2-4 network:

net = feedforwardnet(2,'trainlm');
net = configure(net,randn(4,10),randn(4,10));
inp = randn(4,1);
out = sim(net, inp);

How can I then find the output of the hidden layer?

Regards,

Martijn

Answer 1

[Greg Heath]

The easiest way to obtain the hidden layer output of a I-H-O net is to just use the weights to create a net with no hidden layer with topology I-H.

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Martijn Onderwater]

Ah, got it. I calculated the activations myself, but I missed the fact that MNNT pre- and postprocesses data. See this post for a solution.

Answer 3

[HunterE]

You can also use genFunction to generate a .m file which should exactly reproduce the model in your net object. Then you can edit the resulting .m file to cause it to return the activations. For larger networks this is more practical than re-coding it yourself.

However, is there really no easier way to access the hidden layer activations??? If so this is a serious oversight!!