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
{{ message }}
This repository was archived by the owner on Aug 15, 2024. It is now read-only.
Notes, like the following, deserve to live somewhere:
Conx builds models between any two banks that have fully activated incoming pathways.
Therefore, Network.propagate_from() won't work if you attempt to propagate through a merge (concat, add, mult, etc) layer.
It also doesn't give you multiple pathways when they exist. For example, if you have two pathways between banks "input" and "hidden5", it will just give you the last one.
Conx could also be enhanced to allow propagations that involve multiple input/output banks. I think one can do that manually now, but requires you to hand-wire the model using connections between banks.
Notes, like the following, deserve to live somewhere:
Conx builds models between any two banks that have fully activated incoming pathways.
Therefore, Network.propagate_from() won't work if you attempt to propagate through a merge (concat, add, mult, etc) layer.
It also doesn't give you multiple pathways when they exist. For example, if you have two pathways between banks "input" and "hidden5", it will just give you the last one.
Conx could also be enhanced to allow propagations that involve multiple input/output banks. I think one can do that manually now, but requires you to hand-wire the model using connections between banks.