tha meaning of delay in neural net time series

Hi All,

I want to be sure about the delay in time series meaning For example, when the I write the input delay is 2, and the feedback delay is 3 does that mean its change from 1:2, 1:3? And the other question is can I take the delay in NAREXNET as an intervals? (i.e. 1:2:4)?

Thank you

[Greg Heath]

SPELLING:

 NARXNET   (NOT NAREXNET)

Answer 1

[Greg Heath]

 net = narxnet(ID,FD,H)

 ID is a row vector of NONNEGATIVE, INCREASING BUT NOT
 NECESSARILY CONSECUTIVE, INTEGERS  

 FD is a row vector of POSITIVE , INCREASING BUT NOT NECESSARILY 
CONSECUTIVE, INTEGERS 

 ID and FD do not have to have the same length or integers
 in common

 ID = 2 IS NOT THE SAME AS ID = 1:2 = [ 1 2 ]

 ID = [ 1:2:4 ] = [ 1 3 4 ]

 If ID = [ 0:2:4 ] and FD = [ 1:3 ], then

 y(t+4) = f( x(t+4), x(t+2), x(t), y(t+3, y(t+2), y(t+1) )

Hope this helps,

Thank you for formally accepting my answer

Greg

[Lilya]

Dr. Heath excuse me, but I want to be sure about the value of MSE00 I run the command MSE00 = mean(var(t',1)) to normalized the data the result is 196.8228 t [1 8761] and N= 8761 is it acceptable MSE00?

[Greg Heath]

It has to be acceptable. It is just the average variance of your target variables.

The important point is that it is the minimum MSE that can be achieved with the NAIVE constant output model, that yields the same constant output regardless of input. (The next best model is a linear model)

With a little thought you can prove that if the output is constant, regardless of input, then the constant that minimizes MSE is just the mean of the target matrix and the minimum MSE is just the average target variance.

Therefore, it is a very appropriate reference for normalization.

Also note that if the target columns are standardized (zero-mean/unit-variance) via zscore or mapstd, then MSE00 = 1.

Hope this helps.

Greg

Answer 2

[Fatma HM]

Hi All,

I want to know how I can find the size of the real and the size of the estimated in Neural networks when i have input, output and error ??

GREG:

Real and estimated what?