NARX input delays what exactly do they mean?

Hi,

Im trying to create a neural network to time series prediction with a NARX net. My question is about the tap delays on this network.

Lets say Im using input delays 1:2 and feedback delays 1:2 with inputs X and targets T does this mean I am using the set { x(i-1),x(i-2),t(i-1),t(i-2) } to predict t(i)?

where my X = [x(1),x(2),x(3),….x(n)] and T = [t(1),t(2),t(3),…t(n)]

thanks.

Answer 1

[Greg Heath]

Yes, for i >=3.

Also: note that an input delay of 0 is allowable but a feedback delay of 0 is not.

Thank you for formally accepting my answer

Greg

[Charles]

Thank you I believe I may have asked a question of similar nature today,so this helps. If you remove the input delay does this in effect tell the network to predict a step ahead? Is this why, output 'y' is short in length?