Neural Network Time Series tool

Hello folks,

I want to setup a Neural network to predict some values, I look the Matlab guide for neural network and I find this example:

y = laser_dataset;
y = y(1:600);

ftdnn = timedelaynet([1:8], 10);
ftdnn.trainParam.epochs = 50;
ftdnn.divideFcn = '';
p = A(9:end);
t = A(9:end);
Pi = A(1:8);
ftdnn = train(ftdnn,p,t,Pi);

yp = ftdnn(p,Pi);
e = gsubtract(yp,t);
rmse = sqrt(mse(e))

My question is, how can I see the next predicted value? Or how can I calculate the predicted output data of the neural network?

The manual is a bit confusing because it only indicates how to determine the prediction error, but how could use this RMSE to know the value?

You see, this is a It is a valid question, because I have found a lot of confusion over the Internet on such networks without clear and objective answers.

Another question I have is, I can use the same network (FTDNN) using small samples of input data, say 20 or 30 inputs, providing the output 21 or 31?

Thanks a lot for your help!

[geofo]

The question is still not answered. A neural network is trained that "A==A" (it can be seen the input vector is identical to the output vector, p==t). The writer of the manual is pleased that the error is small. Indeed one can expect that for any input A the output will be precisely A after this training. However the readers are not pleased at all with this example. How can a useful prediction be made after such a tautologic training?

Answer 1

[Greg Heath]

NARNET is the function for a single series

 help NARNET
 doc NARNET

Search BOTH NEWSGROUP and ANSWERS with

 greg NARNET
 greg NARNET tutorial

Hope this helps.

Thank you for formally accepting my answer

Greg