0

I'm new to Neural Networks and I am trying to train an NN by simply loading two different time series data x and y, which are 300 x 1 vectors. x is the set of values of the predictor variable, and y is the dependent (output) variable. By the way, is this logic true? Or do I need to provide different parts of y ?

Anyway, when I call net = train(feedforwardnet(20),x,y); the training stops immediately to tell me that the performance goal is met. See the picture below :

enter image description here

So I guess I need to change some parameters, like the error threshold or something, but I don't know what, so I am seeking for an explanation for what I'm doing wrong.

Thanks for any help !

P.S.

Even though x and y are different, they are following a similar trend, so they are correlated, that may be a clue for the issue. I have also tried normalizing the values, still nothing changes.

jeff
  • 13,055
  • 29
  • 78
  • 136
  • 1
    I will leave the question of validity of your data to someone else. But if you have a statement like `net = feedforwardnet(20);` you can change the parameters of net.trainparam to something else to see if the training starts or not. If the data is correct, it could be that the gradient that is being calculated is actually less than the default threshold used in matlab. Could you tell us the range of y ? – user671805 Aug 29 '14 at 21:48
  • y is between 0 and 1. Yes you are right, I suspect it is related to the threshold itself, so how do I change it ? – jeff Sep 05 '14 at 20:33

0 Answers0