spark.nn.components.delays

spark.nn.components.delays#

Submodules#

Classes#

`Delays`	Abstract synaptic delay model.
`DelaysOutput`	Generic delay model output spec.
`NDelays`	Data structure for spike storage and retrival for efficient neuron spike delay implementation.
`NDelaysConfig`	NDelays configuration class.
`N2NDelays`	Data structure for spike storage and retrival for efficient neuron to neuron spike delay implementation.
`N2NDelaysConfig`	N2NDelays configuration class.

Package Contents#

class spark.nn.components.delays.Delays(config=None, **kwargs)[source]#

Bases: spark.nn.components.base.Component, Generic[ConfigT]

Abstract synaptic delay model.

Parameters:: config (ConfigT | None)

abstractmethod reset()[source]#

Resets component state.

Return type:: None

abstractmethod __call__(in_spikes)[source]#

Execution method.

Parameters:: in_spikes (spark.core.payloads.SpikeArray)
Return type:: DelaysOutput

class spark.nn.components.delays.DelaysOutput[source]#

Bases: TypedDict

Generic delay model output spec.

Initialize self. See help(type(self)) for accurate signature.

out_spikes: spark.core.payloads.SpikeArray[source]#

class spark.nn.components.delays.NDelays(config=None, **kwargs)[source]#

Bases: spark.nn.components.delays.base.Delays

Data structure for spike storage and retrival for efficient neuron spike delay implementation. This synaptic delay model implements a generic conduction delay of the outputs spikes of neruons. Example: Neuron A fires, every neuron that listens to A recieves its spikes K timesteps later,

neuron B fires, every neuron that listens to B recieves its spikes L timesteps later.

Init:: max_delay: float delay_initializer: DelayInitializerConfig
Input:: in_spikes: SpikeArray
Output:: out_spikes: SpikeArray

Parameters:: config (NDelaysConfig)

config: NDelaysConfig[source]#

build(input_specs)[source]#

Build method.

Parameters:: input_specs (dict[str, spark.core.specs.PortSpecs])

reset()[source]#

Resets component state.

Return type:: None

get_dense()[source]#

Convert bitmask to dense vector (aligned with MSB-first packing).

Return type:: jax.Array

__call__(in_spikes)[source]#

Execution method.

Parameters:: in_spikes (spark.core.payloads.SpikeArray)
Return type:: spark.nn.components.delays.base.DelaysOutput

class spark.nn.components.delays.NDelaysConfig(__skip_validation__=False, **kwargs)[source]#

Bases: spark.nn.components.delays.base.DelaysConfig

NDelays configuration class.

Parameters:: __skip_validation__ (bool)

max_delay: float[source]#

delays: jax.Array | spark.nn.initializers.base.Initializer[source]#

class spark.nn.components.delays.N2NDelays(config=None, **kwargs)[source]#

Bases: spark.nn.components.delays.base.Delays

Data structure for spike storage and retrival for efficient neuron to neuron spike delay implementation. This synaptic delay model implements specific conduction delays between specific neruons. Example: Neuron A fires and neuron B, C, and D listens to A; neuron B recieves A’s spikes I timesteps later,

neuron C recieves A’s spikes J timesteps later and neuron D recieves A’s spikes K timesteps later.

Init:: units: tuple[int, …] max_delay: float delays: jnp.ndarray | Initializer
Input:: in_spikes: SpikeArray
Output:: out_spikes: SpikeArray