NARX test set closed loop

Hi all, Does NARX's test set use a closed loop network automatically in it's train(net,x,t,xi,ai) function? Since principally for the test dataset the network is just fed with input and the target is withheld from the network for testing. Thanks

Answer 1

[Greg Heath]

You seem confused:

 1. OL trn, val and tst subsets are automatically 
    tested by TRAIN.
 2. The OL configuration is then explicitly 
    converted to CL by CLOSELOOP.
 3. CL performance of the trn, val and tst subsets 
    is then obtained after the conversion.

You can see some recent examples by searching in BOTH NEWSREADER and ANSWERS using

 greg narxnet closeloop

Hope this helps.

Thank you for formally accepting my answer

Greg

[Tu Yuan Neoh]

Thanks for that G. On a side note, could you explain to me in simple terms how removedelay(net) really works? Matlab's doc removedelay says that it decreases input delays and increases output feedback delays. If I don't closeloop before removedelay, shouldn't it be a target feedback instead of an output feedback. And how does shifting the delays enable step ahead prediction?

[Greg Heath]

After reading the documentation on removedelay, I decided it's a lot of bullfeathers. So I don't use it and do not recommend it.

Hope this helps.

Greg

PS if you wish to try it I suggest more training after the delays are reduced.