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
The implementation of train_step is rather generic and is more or less boilerplate code that the models have to write (for both built in models and externally defined models)
We should be able to define a default train_step that can be overridden by a users model if needed.
It makes sense to include a default implementation in the @fibad_model decorator as one of the methods that is automatically added to the annotated class.
This approach also gives us a pathway to getting accuracy and loss values (using the built in metrics provided by pytorch-ignite) without requiring the user to define or calculate those. And then we can push those to TensorBoard as well.
The text was updated successfully, but these errors were encountered:
The implementation of
train_step
is rather generic and is more or less boilerplate code that the models have to write (for both built in models and externally defined models)We should be able to define a default
train_step
that can be overridden by a users model if needed.It makes sense to include a default implementation in the
@fibad_model
decorator as one of the methods that is automatically added to the annotated class.This approach also gives us a pathway to getting accuracy and loss values (using the built in metrics provided by pytorch-ignite) without requiring the user to define or calculate those. And then we can push those to TensorBoard as well.
The text was updated successfully, but these errors were encountered: