Albin Soutif

> I agree about the `nc_first_task` option, we should also have it for multi-task scenarios. > > Your snippets seems wrong. Instead of modifying the experiences in place, it's easier...

fixed in #1285

> This is an issue indeed. I agree with your solution of disabling adaptation. > > Another issue that we have is that the model adaptation must be called on...

@AntonioCarta I might work on that next. I think we should indeed implement a load_state_dict function for the dynamicmodules that automatically adapts the module. That would simplify a lot of...

Maybe we have to wait for this issue: https://github.com/pytorch/pytorch/issues/75287 to be solved. It seems like these load_state_dict pre hooks would be the way to go, but they are implemented as...

@AntonioCarta How would you like to manage the adapt() and adaptation() function ? I wanted to keep the name of the "adaptation" also to keep backward compatibility but now it's...

> > @AntonioCarta How would you like to manage the adapt() and adaptation() function ? I wanted to keep the name of the "adaptation" also to keep backward compatibility but...

> > Made a few clarification changes in IncrementalClassifier and Multiheadclassifier by replacing occurences of adaptation() implementation by train_adaptation and eval_adaptation, where if the eval_adaptation needs to be the same...

@AntonioCarta On my side everything is ok. I was thinking maybe to include an example on how to use dynamic modules outside and inside of a strategy (call to recursive_adaptation...

Ok, should I still try to change GenericForgettingExperience so that it works with Online Scenarios when task boundaries are provided or do you think the structure of all forgetting metrics...