You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
At least for Hessian Free, it would be very useful to automatically save the best performing epoch on a validation set as the network to use. Currently, we are watching the actual test set, so it would be cheating to use it to determine when to stop. Additionally, we need a way to save the whole model each time as the 'final' model, when the error on validation is lowest. This would require bringing the save functionality inside the Hessian Training function.
The text was updated successfully, but these errors were encountered:
At least for Hessian Free, it would be very useful to automatically save the best performing epoch on a validation set as the network to use. Currently, we are watching the actual test set, so it would be cheating to use it to determine when to stop. Additionally, we need a way to save the whole model each time as the 'final' model, when the error on validation is lowest. This would require bringing the save functionality inside the Hessian Training function.
The text was updated successfully, but these errors were encountered: