spynnaker.pyNN.models.abstract_models package

Module contents

class spynnaker.pyNN.models.abstract_models.AbstractAcceptsIncomingSynapses

Bases: object

Indicates an application vertex that can be a post-vertex in a PyNN projection.

Note

See verify_splitter()

abstractmethod clear_connection_cache() → None

Clear the connection data stored in the vertex so far.

abstractmethod get_connections_from_machine(app_edge: ProjectionApplicationEdge, synapse_info: SynapseInformation) → ConnectionsArray

Parameters:

• app_edge – The edge for which the data is being read

• synapse_info – The specific projection within the edge

Returns:

The connections from the machine post-run.

abstractmethod get_synapse_id_by_target(target: str) → int | None

Parameters:

target – The name of the synapse

Returns:

The ID of the synapse with this name

abstractmethod set_synapse_dynamics(synapse_dynamics: AbstractSynapseDynamics) → None

Set the synapse dynamics of this vertex.

Parameters:

synapse_dynamics

verify_splitter(splitter: AbstractSpynnakerSplitterDelay) → None

Check that the splitter implements the API(s) expected by the SynapticMatrices

Any Vertex that implements this API should override ApplicationVertex.splitter method to also call this function.

Parameters:

splitter – the splitter

Raises:

PacmanConfigurationException – if the splitter is not an instance of AbstractSpynnakerSplitterDelay

class spynnaker.pyNN.models.abstract_models.AbstractHasDelayStages

Bases: object

Indicates that this object (an application vertex) has delay stages that are used to increase the space required for bitfields in spynnaker.pyNN.utilities.bit_field_utilities.get_estimated_sdram_for_bit_field_region()

abstract property n_delay_stages: int

The maximum number of delay stages required by any connection out of this delay extension vertex.

class spynnaker.pyNN.models.abstract_models.AbstractMaxSpikes

Bases: object

Indicates a class (a MachineVertex) that can describe the maximum rate that it sends spikes.

The SynapticManager assumes that all machine vertexes share the same synapse_information will have the same rates.

abstractmethod max_spikes_per_second() → float

Returns:

The maximum expected number of spikes per second.

abstractmethod max_spikes_per_ts() → float

Returns:

The maximum expected number of spikes per timestep.

class spynnaker.pyNN.models.abstract_models.AbstractNeuronExpandable

Bases: object

Indicates a class (a MachineVertex) that has may need to run the neuron expander APLX.

abstractmethod gen_neurons_on_machine() → bool

True if this vertex should be generated on the machine.

Note

The typical implementation for this method will be to ask the neuron data

Returns:

True if this vertex can generate its data on machine

abstract property neuron_generator_region: int

The region containing the parameters of neuron expansion.

abstractmethod read_generated_initial_values(placement: Placement) → None

Fill in any requested initial values.

Parameters:

placement – Where the data is on the machine

class spynnaker.pyNN.models.abstract_models.AbstractSynapseExpandable

Bases: object

Indicates a class (a MachineVertex) that has may need to run the synapse expander APLX.

Note

This is not implemented by the DelayExtensionMachineVertex, which needs a different expander APLX

abstract property bit_field_size: int

The amount of bit field data to be generated. This is used to calculate the timeout of the execution.

abstract property connection_generator_region: int

The region ID containing the parameters of synaptic expansion.

abstractmethod gen_on_machine() → bool

True if the synapses of this vertex should be generated on the machine.

Note

The typical implementation for this method will be to ask the app_vertex’s synapse_manager

Returns:

True if this vertex can generate its data on machine

abstract property max_gen_data: int

The maximum amount of synaptic data to be generated. This is used to calculate the timeout of the execution.

abstractmethod read_generated_connection_holders(placement: Placement) → None

Fill in the connection holders.

Parameters:

placement – Where the data is on the machine

class spynnaker.pyNN.models.abstract_models.ColouredApplicationVertex(label: str | None = None, max_atoms_per_core: int | Tuple[int, ...] | None = None, splitter: AbstractSplitterCommon | None = None)

Bases: ApplicationVertex

Indicates an application vertex that implements n_colour_bits.

Parameters:

• label – The optional name of the vertex.

• max_atoms_per_core – The max number of atoms that can be placed on a core for each dimension, used in partitioning. If the vertex is n-dimensional, with n > 1, the value must be a tuple with a value for each dimension. If it is single-dimensional the value can be a 1-tuple or an int.

• splitter – The splitter object needed for this vertex. Leave as None to delegate the choice of splitter to the selector.

abstract property n_colour_bits: int

The number of colour bits sent by this vertex.

Assumed 0 unless overridden

class spynnaker.pyNN.models.abstract_models.HasShapeKeyFields

Bases: object

Indicates a source that has keys in fields for each dimension of the source.

abstractmethod get_shape_key_fields(vertex_slice: Slice) → List[Tuple[int, int, int, int]]

Parameters:

vertex_slice – The slice of the source vertex

Returns:

Get the fields to be used for each dimension in the shape of the given source vertex slice, as a list of start, size, mask, shift values in the order of the fields.

class spynnaker.pyNN.models.abstract_models.HasSynapses

Bases: object

API for getting connections from the machine.

abstractmethod get_connections_from_machine(placement: Placement, app_edge: ProjectionApplicationEdge, synapse_info: SynapseInformation) → Sequence[NDArray]

Parameters:

• placement – Where the connection data is on the machine

• app_edge – The edge for which the data is being read

• synapse_info – The specific projection within the edge

Returns:

The connections from the machine for this vertex.

class spynnaker.pyNN.models.abstract_models.ReceivesSynapticInputsOverSDRAM

Bases: AbstractSupportsSDRAMEdges

An object that receives synaptic inputs over SDRAM.

The number of neurons to be sent per synapse type is rounded up to be a power of 2. A sender must send N_BYTES_PER_INPUT of data for each synapse type for each neuron, formatted as all the data for each neuron for the first synapse type, followed by all the data for each neuron for the second, and so on for each synapse type. Each input is an accumulated weight value for the timestep, scaled with the given weight scales.

N_BYTES_PER_INPUT = 2

abstract property n_bytes_for_transfer: int

The number of bytes to be sent over the channel. This will be calculated using the above numbers, but also rounded up to a number of words, and with the number of neurons rounded up to a power of 2.

abstract property weight_scales: ndarray[tuple[Any, ...], dtype[floating]]

A list of scale factors to be applied to weights that get passed over SDRAM, one for each synapse type.

class spynnaker.pyNN.models.abstract_models.SendsSynapticInputsOverSDRAM

Bases: AbstractSupportsSDRAMEdges

An object that sends synaptic inputs over SDRAM.

abstractmethod set_sdram_partition(sdram_partition: SourceSegmentedSDRAMMachinePartition) → None

Set the SDRAM partition. Must only be called once per instance.

Parameters:

sdram_partition – The SDRAM partition to receive synapses from

class spynnaker.pyNN.models.abstract_models.SupportsStructure

Bases: object

Indicates an object that supports the setting of a PyNN structure.

abstractmethod set_structure(structure: BaseStructure) → None

Set the structure of the object.

Parameters:

structure – The structure to set