[NARX] How can I interpret the input weight with delay

Hi all, I have created a NARX neural network model using matlab with 4 exogenous variables, 2 delays and 3 hidden neurons.

After training,I want to retrieve the weights of each input and the corresponding delay input. By using net.IW{1}, i have returned the following output: >> net.IW{1}

ans =

      0.0447   -0.2305   -0.6253    0.5424   -0.2860   -0.0664   -0.4632   -0.8809
      0.0167   -0.1856   -0.1209    0.2211   -0.2072   -0.2291   -0.1980   -0.4857
      0.0333   -0.1057    0.4207   -0.1289   -0.0046   -0.1781    0.0888    0.0221

which is 3X8.

So here is the question: for the 1st neuron, how can I interpret the 8 numbers? which of the 2 are correspond to the 1st exogenous variable? is it (0.0447 and -0.2305) or (0.0447 and -0.2860)

Answer 1

[Greg Heath]

To be consistent with matrix multiplication, it has to be (0.0447 and -0.2860)

 h(t) = b1 + IW*[ x(:,t-1) ; x(:,t-2)]; % t = 3,4,...
 y(t) = b2 + LW*h(t);

Hope this helps.

Thank you for formally accepting my answer

Greg

[fransisca fransisca]

Hi I have 6 exogenous inputs with 2 delays and 6 hidden nodes.

I got matrix 12x6. Shouldn't it have to 18x6?

since the input will be x(t), x(t-1), x(t-2)?

Really appreciate your answer. Thank you.