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 idea of the train state is that it includes everything you need for training.
I think it is reasonable to get to the parameters through train state only, because using params as an additional attribute might seem confusing then.
I actually would prefer it how it is right now.
I don't really mind either way, I would suggest that if you make the param then you need to be absolutely sure that it gets updated automatically when the model state is updated, i.e that it is a pointer, which I do not know off the top of my head. If you can ascertain this and you can make that work nicely then I would say do as you please. The params are fundamental even if the train state goes which for other models it might.
Instead of writing something like
network.model_state.params
we can add a property to the class for convenienceThis is just a tracker for a future PR until #49 is merged
The text was updated successfully, but these errors were encountered: