Print

Print


On May 6, 2017, at 9:34 PM, Chris Johnson <[log in to unmask]> wrote:

> I do have a question.  If this question belong elsewhere, please point me there and there I will ask it. 

It probably belongs in another venue.  I'd suggest the GP Mailing List.

What you're talking about is overfitting and generalization.  Originally the GP symbolic regression crowd were just interested in fitting an expression; but generalization (thus not overfitting) is desirable as it would be in any machine learning context.

ECJ has testing/training options for generalization in its benchmark.params file, which you might check out.   But as to *how* to avoid overfitting, that's really more of a research question and not an ECJ one per se.

Sean