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
With a few extra tweaks, I think the new arbitrary input/output scheme could support recurrent architectures. Would just need to keep the intermediate DataSet object (i.e. containing all saved outputs) around during fitting, which should probably be done anyway to reduce overhead, and add a way to specify some kind of default if the output doesn't exist (e.g. on first evaluation, if receiving input from a later layer that hasn't been evaluated yet).
The text was updated successfully, but these errors were encountered:
With a few extra tweaks, I think the new arbitrary input/output scheme could support recurrent architectures. Would just need to keep the intermediate DataSet object (i.e. containing all saved outputs) around during fitting, which should probably be done anyway to reduce overhead, and add a way to specify some kind of default if the output doesn't exist (e.g. on first evaluation, if receiving input from a later layer that hasn't been evaluated yet).
The text was updated successfully, but these errors were encountered: