LBFGS optimizer not available
Closed this issue · 4 comments
pradhyumna85 commented
Is there any way to use LBFGS optimizer in SciANN ?
ehsanhaghighat commented
In scimodel, typye ‘optimizer=“scipy-****” Replace **** with your desired optimizer from scipy library On Aug 23, 2022, at 10:36 AM, Pradyumna Singh Rathore ***@***.***> wrote:
Is there any way to use LBFGS optimizer in SciANN ?
—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>
pradhyumna85 commented
Okay, that was helpful. But now the issue I observe is even though I am giving say 100 epochs in model.train, still the training stops training early after no loss change after some iterations, i.e., doesn't go till the epochs specified and I am not using any early stopping.
ehsanhaghighat commented
Yeah, scipy does its own thing although I pass epochs (max iterations). If you want to customize it, change the optimizer.py inside utils to follow your needs.
… On Aug 24, 2022, at 2:26 AM, Pradyumna Singh Rathore ***@***.***> wrote:
Okay, that was helpful. But now the issue I observe is even though I am giving say 100 epochs in model.train, still the training stops training early after no loss change after some iterations, i.e., doesn't go till the epochs specified and I am not using any early stopping.
—
Reply to this email directly, view it on GitHub <https://github.com/sciann/sciann/issues/47#issuecomment-1225465542>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHMBIZ6JQ5DBMMX6PQE5BLDV2XTGHANCNFSM57MLJDCQ>.
You are receiving this because you commented.
pradhyumna85 commented
Ok Ehsan, thank you for the prompt response. I am closing this issue, but if you have some resources on how to override this behavior of scipy then I'll be very thankful to you.🙂