v1.1.0 Change Logs
- Added supports for multiple layers.
- Renamed
sparsity_constraints
assparsity_masks
. - Renamed
positivity_constraints
asei_masks
. - Added
plasticity_masks
, defines whether a connection is fixed throughout the training. - Added
synapse_growth_masks
, defines whether a node could grow new synapses (whether an edge with zero value could change to nonzero).
Todos:
- Add custom constraint wrapper.
- Change
visualize()
in thenn4n.structure
toplot_masks()
. - Change
masks()
in thenn4n.structure
toget_masks()
. - Change the loss to functional forms. (e.g. function(model) -> loss). It might be better to also have a function that gets the specific model attributes (i.e., an interface cuz the model might change).
- Rename
criterion
toconstraint
. - Change quickstart example to a more complex example.
- Change
BaseStruct
toBaseMask
. - Change top icons to actual links
- The EIRNN example need to be change to the actual example.