By including a moving window of fixed length in the input vector of MLP, is the Back-propagation ANN equivalent to NAR model?

If this is the case, how we can add the moving window? Supposing that the lag is equal to 3, for example:

N= lenght(data);
d=timestep ahead;
input  = data( 1:N-d);   % No transpose; 
target = data( 1+d : N );
MSE00  = var(target',1)  % Reference MSE
net = fitnet;   % default H = 10 
net.divideParam.valRatio = 10/100;
net.divideParam.testRatio = 20/100;
[net tr output error ] = train(net, input, target);
%output = net(input); 
error = target - output;
NMSE = mse(error)/MSE00 % Range [ 0 1 ]
R2   = 1- NMSE

Thanks

Answer 1

[Greg Heath]

1. When you insert code try to make sure it runs.

 N= lenght(data);  % ERROR

 d=timestep ahead; % ERROR

2. Replace TRAIN with ADAPT

Hope this helps.

Thank you for formally accepting my answer

Greg

[coqui]

thank you Greg.

I only have 1 series, I have used FITNET. To continue beyond the original data (for example, 50 points) how I can do it?

[Greg Heath]

I have several posts on predicting data beyond the target region. Let me know if you can't find any of them.