perpdgo
commented
3 years ago In the case of Bayesian regularized neural networks, you can try the effective number of parameters, which is a parameter that ranges between 1 and the total number of parameters to estimate. So what you can do is to try to minimize the difference between the number of parameters to estimate and the effective number of parameters. For a given model, you need to fit the model several times, so that you can obtain the mean of the difference between these two quantities, the effective number of parameters can change from run to run, because the parameters are initialized at random.
Regards.
On Sun, Aug 8, 2021 at 7:38 AM asheetal @.***> wrote:
I hope someone can advice on hyperparameters. I am using brnn in conjuction with caret. Caret provides only a single tuning parameter neurons. However it is really slow to walk through crossvalidation and grid search. Is there is any empirical finding on choice of number of neurons?
hyperparameter+crossvalidation is really making the analysis slow
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/perpdgo/brnn/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC3ARLSMO3BSGVX5KKNLM3LT3Z3EHANCNFSM5BYOMW7A . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .
I hope someone can advice on hyperparameters. I am using brnn in conjuction with caret. Caret provides only a single tuning parameter neurons. However it is really slow to walk through crossvalidation and grid search. Is there is any empirical finding on choice of number of neurons?
hyperparameter+crossvalidation is really making the analysis slow