Prev: ??? Out of memory. Type HELP MEMORY for your options.
Next: Replacing cells that have elements
From: Jorge Pasquotto on 11 Apr 2010 13:23
Dear all,
I don't understand why in some situations the NNToolbox (Matlab 2008a) runs a training session without cross-validating the adjusted bias/weights?
For example, see code below where I am trying to train an Elman-5 network to fit the function y =f(x) = 3x+1.
>> P={1 2 3 4 5 6 7 8 9 10};
>> T={4 7 10 13 16 19 22 25 28 31};
>> S1=5;
>> net=newelm(P,T,S1);
>> net.gradientFcn='calcgbtt';
>> trainInd=(1:7);
>> valInd=(8:9);
>> testInd=10;
>> netdivideFcn='divideInd';
>> net=train(net,P,T);
In this moment I can follow the training process, I see the performance increases (MSE decreases) for 1000 epochs in the training set...however no validation check occurs???
Why?
I would expect that from time to time, the software would show MSE for training and validation sets...however 1000 epochs are elapsed and not validation check occurs. Why?

regards
Jorge
 | 
Pages: 1
Prev: ??? Out of memory. Type HELP MEMORY for your options.
Next: Replacing cells that have elements