"Greg Heath" <heath@alumni.brown.edu> wrote in message <khr3fc$mmk$1@newscl01ah.mathworks.com>...
> Some of the time series posts indicate that there is some confusion regarding the
> purpose, domain and overlap of the three basic timeseries functions. With the hope of reducing that confusion, I have posted posted some of my notes below:
>
> % TIMESERIES REGRESSION AND PREDICTION CATEGORIES
>
> net = narxnet( ID, FD, H ); % The most general openloop timeseries net.
>
> ID Empty Row vector or Row vector of increasing NONNEGATIVE integers
> FD Empty Row vector or Row vector of increasing POSITIVE integers
> H Empty Row vector or Row vector of POSITIVE integers
>
> ID Values of input delays
> FD Values of output feedback delays
> H Number of nodes per hidden layer
>
> y(t) = f( x( t  id : t ), y( t  fd : t  1 ), H ), id >= 0, fd >=1
>
> SPECIAL CASES
>
> 1. Regression: min(ID) = 0
>
> 2. Prediction: min(ID) > 0
>
> 3. LINEAR: H = []
>
> 4. TIMEDELAYNET: FD =[]
>
> net = narxnet( ID, [], H);
>
> = timedelaynet( ID, H );
>
> y(t) = f( x( t  id : t ) , H )
>
> 5. NARNET: ID = []
>
> net = narxnet( [], FD, H );
>
> = narnet( FD, H );
>
> y(t) = f( y( t  fd : t  1 ), H )
Although, it makes sense that TIMEDELAYNET and NARNET should be available
as special cases of NARXNET, the current MATLAB implementation does not allow those special cases.
BUMMER!
Sorry, for the misinformation.
Greg
