How we can define the number of expansion of first input by using trigonometric functional link artificial neural network?

In trigonometric functional link artificial neural network, each input sample is expanded to N sine terms, N cosine terms plus the sample itself. How we can define N ?

Answer 1

[Greg Heath]

From Fourier Series

 N = T/dt

 T = length of sample

 dt = sampling time

Hope this helps.

Thank you for formally accepting my answer

Greg

[coqui]

Thank you Greg. But,I need more explanation. I have used five inputs for a period running from 2002 to 2014 (3300 daily observations) in order to predict next index. How can compute N? Is N equals to 3300/3300=1 in our case?

thanks

[Greg Heath]

How many days do you want to predict ahead ? D? Then N= 3300*D. However, I'm sure you don't need 3300 per day. Therefore you need to figure out how to subsample WITHOUT LOSING ESSENTIAL INFO.

If you do it by trial and error, try recursively reducing the size by 2.

Hope this helps.

Greg

[coqui]

I need to predict one day ahead by using functional link artificial neural network with hyperbolic tangent transfer function in output layer. what about the use of three expansion of first input Y1, i.e. are y1=Y1, y2=cos(πY1), y3=sin(πY1)???

[Greg Heath]

1. I AM NOT FAMLIAR WITH THE FUNCTIONAL LINK NET AND DON'T SEE THE ADVANTAGE OF ADDING THE FOURIER TERMS.

2. WHAT YOU HAVE WRITTEN ABOVE FOR Y2 AND Y3 MAKES ASOLUTELY NO SENSE TO ME. THEY ARE NOT TERMS IN THE EXPANSION OF Y1.