Open dsblank opened 3 weeks ago
It looks like we only need to update the parts of aitk.network that dived into the internals of tensorflow. These are the parts of the code that touch "nodes". Rewriting them should allow aitk to work without replying on tensorflow 2.15.
The only limitation is that any aitk.network is going to require a connection pairing between all in/out pairing of layers. This makes it harder to Network(model)
but that wasn't a major focus for us anyway. You'd now need to Network(model, connections)
to avoid accessing TF internals.
Found one other change: layer.input
was changed to layer._input_tensor
Can now build a SimpleNetwork. One more change: layer.output_shape
has been removed.
I will list the issues here in refactoring aitk version 2, to work with the latest keras in tensorflow 2.17.
I'll start a new branch, aitk-3, to fix the issues.