HNSLayer

Bases: tf.keras.layers.Layer

Hub-and-Spokes RNN Layer.

Unrolling the HNSCell for a fixed number of time steps.

See Rogers et. al., 2004 for more details.

Parameters:

Name	Type	Description	Default
tau	float	Time constant of the time-averaging.	required
hub_name	str	Name of the hub layer.	required
hub_units	int	Number of units in the hub layer.	required
spoke_names	List[str]	Names of the spoke layers.	required
spoke_units	List[int]	Number of units in each spoke layer. Must be the same length as spoke_names.	required

build(input_shape)

Build the HNSCell.

call(inputs=None, return_internals=False)

Forward pass: unroll the HNSCell for a fixed number of time steps.

The number of time steps is determined by axis 1 in the inputs.

Parameters:

Name	Type	Description	Default
inputs	Dict[str, tf.Tensor]	Inputs to the spokes (name as key). Assumes input is 0 if not given.	None
return_internals	bool	Whether to return intermediate inputs to each connection.	False

Returns:

Type	Description
Dict[str, tf.Tensor]	Dict[str, tf.Tensor]: Activations of the hub and spokes, with layer names as keys. If return_internals is True, also returns intermediate inputs to each connection.