Creating A Layer Recurrent Neural Net with more than 1 Hidden State.

Hello,

Whenever I create a Layer Recurrent Neural Network, I do the following :

laynet = layrecnet(3); 3 Inputs... 3 Neurons (I hope this is correct). After this, if I do laynet.numLayers = 3 this only creates an extra output layer.

What I need to do is create a layrecnet with 2 hidden states with 3 inputs and 1 output. How do I exactly achieve this?

Answer 1

[Greg Heath]

1. What is the definition of a hidden state?...the output of a hidden layer?

2. Why do you think you need a layrecnet instead of a narxnet?

3. Why do you think you need 2 hidden layers?

4. To see what a creation command yields, just use the view(net) command

5. net = layrecnet(1:2,[ 5 5 ]) % yields 2 hidden states

6. If

[ 3 N ] = size(X)
[1 N ] = size(T)

then either of

 net = configure(net,X,T);

or

 net = train(net,X,T);

will yield the correct I/O dimensions

Thank you for formally accepting my answer

Greg

[Aditya Tirodkar]

I'll accept in in a minute. Please, what does the 1:2 do? And the 5 as well?

[Greg Heath]

See

   help layrecnet

and

   doc layrecnet

for details.

Then see the configuration via

 net = layrecnet(1:2, [5 5])

 view(net)

The short answer is

 There are two 5-dimensional hidden states with 2-dimensional feedback signals to the input of the hidden layers.