Re:Introduction to Artificial Neural Networks – Part 1
by howeman on Saturday December 07, 2013 @12:17PM
(#6866892)
I think that's basically it. If you want to do a lot of work you can do somethings with cross validation (see which parameters perform the best), but that's a form of automatic try/fail
[ Parent
| on HN ]
Re:Introduction to Artificial Neural Networks – Part 1
by pilooch on Saturday December 07, 2013 @12:28PM
(#6866923)
hyperparameter search is a solution. See http://jmlr.org/papers/v13/bergstra12a.html and http://jaberg.github.io/hyperopt/