'preparets' command in Matlab Ver 7.9.0 (R2009b)

I am creating a narx network and am in the last phase of training the network. Can someone help me to address the following please;

I can't find the command for preparing inputs and targets (preparets) in R2009b. I worked previously on a later version and I could find. Also after training and simulating the network, which are the exact codes for predicting a series e.g. if the data you had was up to t=500 and you used the data e.g. up to 300 for training and up to 100 each for validation and testing. What is the exact code for extending the prediction say to get an output y at t=600?

Thanks in advance

[FRANCISCO]

Hi, could you answer this question?, Is that I am now in the same situation. Many thanks.

Answer 1

[Greg Heath]

1. Run all of the original data through the open loop net. Tabulate and plot the resulting OL error. You may want to color code the separate training, validation and testing regions.

2. Close the loop to create netc.

3. Run all of the original data through the CL net. Tabulate and overlay the plot of the CL error on the OL error plot.

4. Run different windows of the original data through the CL net to determine how long it can be run with acceptable errors provided the initial feedback delays are obtained from target data.

Hope this helps.

Thank you for formally accepting my answer

Greg

[FRANCISCO]

Well, I've done from the testing of the accuracy in openloop (whose results are in the previous post) was this:

1 - I caught openloop errors and graphically represent

2 - Then I closed the loop and with the training data / validation and test, I ran it, I figured the mistakes and represent them in a graph.

3 - Then graph openloop errors and errors closeloop in the same graph and compared (being very similar errors and closeloop openloop).   The next step that you have shown me not quite understand: "Run different windows of the original data through the net to determine CL how long it can be run with acceptable errors provided the initial feedback delays are Obtained data from target." How did I do? simply taking input data ranges and targets and performing closeloop?

My last question is that since I've been thinking for some time: If I'm on day 08/03/2013 and let me know the next 10 days, knowing only the inputs of the day 08/03/2013, how would the prediction?. 'd Use the outputs predicted, but as modify the command matlab PREPARETS to let me closeloop similar? Thank you very much.

[Greg Heath]

Yes. Feedback closeloop prediction errors will eventually accumulate and cause unacceptable performance degradation. So the question is,"If I start at an arbitrary specified time and initialize with target data, how long can I rely on my predictions?".

That is the answer to both questions.

Answer 2

[Greg Heath]

Using the commands type closeloop and type preparets, I found the copyright dates of 2010 and 2010/2011, respectively. So if you have 2009b you probably don't have access to them. Are you able to use someone else's version?

You don't need preparets to design an openloop timeseries. Just use delayed versions of input and target data as additional dimensionsof the input matrix.

Most likely, there is a feedback net or two that you can substitute for the closeloop option. Otherwise you wold have to build a custom net.

Check the 2009b documetation for timeseries and feedback nets.

Hope this helps.

Thank you for formally accepting my answer.

Greg

[FRANCISCO]

Now I have installed the 2012 version of matlab, and I modified the code, I would like to know how to perform 10 iterations from today's data, ie, introducing the 9th data, get the next 10 days. As I said earlier predicted using outputs, but do not know how to do ten iterations taking inputs closeloop nothing today. Openloop accuracies are posted in previous post, which I would like to know how to make these ten iterations. Thank you very much.