search

PrincetonUniversity / PsyNeuLink

A block modeling system for cognitive neuroscience
https://psyneulink.org
Apache License 2.0
86 stars 31 forks source link

Default Learning Rates #2015

Open tylergiallanza opened 3 years ago

tylergiallanza commented 3 years ago

Currently, the default learning rate for the add_backpropagation_learning_pathway function of Composition (learning_rate=0.05) is different than the default learning rate for AutodiffComposition (learning_rate=0.001).

There may be good reasons for this discrepancy, but it's a bit unnatural when trying to compare the results of learning in a Composition with and without autodiff. I think it may be beneficial to make these the same value to avoid any confusion.

NOTE: this should be labeled as a feature request and not a bug - I'm not sure how to do that, so sorry for any confusion

jdcpni commented 3 years ago

Thanks. Agreed.

jdc

On Apr 21, 2021, at 4:42 PM, tylergiallanza @.**@.>> wrote:

Currently, the default learning rate for the add_backpropagation_learning_pathway function of Composition (learning_rate=0.05) is different than the default learning rate for AutodiffComposition (learning_rate=0.001).

There may be good reasons for this discrepancy, but it's a bit unnatural when trying to compare the results of learning in a Composition with and without autodiff. I think it may be beneficial to make these the same value to avoid any confusion.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/PrincetonUniversity/PsyNeuLink/issues/2015, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ACSHKGD2M7KXYJF2PR3T55LTJ42EPANCNFSM43LDHPIA.

kmantel commented 3 years ago

Which should be the default (or should both be changed to something else)?