Explanation inputDelays and feedbackDelays in NARX

I need to verify that I am interpreting correctly how to set up delays in a narx network.

Everything that I found in the examples and documentation always start with 0 or 1 like:

inputDelays=1:2
feedbackDelays=1:2

Right now, I have my delays set as:

inputDelays=7:17
feedbackDelays=7:17

and my goal is to predict data like in the schematics below:

I need to predict "to" using data starting at 7 time steps in the past and use data from 10 timesteps to make the prediction.

Are my delays set correctly?

Answer 1

[Samatha Aleti]

Hi,

The “inputDelays”, “feedbackDelays” defines the data that is used to predict output of current time series. If input delay is “i” it implies that the input value at “i” time steps ago(past) is used to predict current output and the same applies for feedback delay.

The condition to define “inputDelays”, “feedbackDelays” are that both must be row vectors with non-negative values in an increasing order. These values need not be consecutive, need not start from a specific number. Also, both these vectors are independent in terms of length and values.

Hope this helps!