neural network ntstool trains too fast!

Hi, i have used sample datasets provided from matlab to simulate a NARX time series prediction model. When i clicked on train, the training stops after few iterations.. It is due to the maximum value of validation check - 6.

What is the best practice if one encounter such datasets? Is it recommended to change the 'divideblock' to ' ' so that validation check will not disrupt the training process and achieve good results.

Answer 1

[Greg Heath]

Validation stopping prevents the net from performing badly on nontraining data.

If the all 3 error rates tr.best_perf, tr.best_vperf and tr.best_tperf are not sufficiently low compared to the average target variance, design another net starting with a different random number state.

The best approach is to train multiple nets and choose the one with the lowest validation set error.

For examples searh using

 greg Ntrials

Hope this helps

Thank you for formally accepting my answer

Greg