graph.LinearWeights

class graph.LinearWeights(input_nodes: rockpool.graph.graph_base.SetList[GraphNode], output_nodes: rockpool.graph.graph_base.SetList[GraphNode], name: str, computational_module: Any, weights: Union[float, numpy.ndarray, torch.Tensor], biases: Optional[Union[float, numpy.ndarray, torch.Tensor]] = None)

Bases: rockpool.graph.graph_base.GraphModule

A GraphModule that encapsulates a single set of linear weights

Attributes overview

biases	The biases (Nout,) encapsulaed by this module
weights	The linear weights (Nin, Nout) encapsulated by this module

Methods overview

__init__(input_nodes, output_nodes, name, ...)
add_input(node)	Add a GraphNode as an input source to this module, and connect it
add_output(node)	Add a GraphNode as an output of this module, and connect it
clear_inputs()	Remove all GraphNode s as inputs of this module
clear_outputs()	Remove all GraphNode s as outputs of this module
remove_input(node)	Remove a GraphNode as an input of this module, and disconnect it
remove_output(node)	Remove a GraphNode as an output of this module, and disconnect it

__init__(input_nodes: rockpool.graph.graph_base.SetList[GraphNode], output_nodes: rockpool.graph.graph_base.SetList[GraphNode], name: str, computational_module: Any, weights: Union[float, numpy.ndarray, torch.Tensor], biases: Optional[Union[float, numpy.ndarray, torch.Tensor]] = None) → None

classmethod _convert_from(mod: rockpool.graph.graph_base.GraphModuleBase) → rockpool.graph.graph_base.GraphModuleBase

Convert another GraphModule to a GraphModule of this specific subclass

You should override this method in your subclass, to include conversion rules from other graph module classes to your specific subclass.

If you do not provide conversion rules to your specific subclass then it will not be possible to map other GraphModule subclasses to your subclass.

Parameters

mod (GraphModule) – A GraphModule or GraphModule subclass object to convert to an object of the specific subclass.

Returns

A converted GraphModule subclass object, of the specific subclass on which this method was called.

Return type

GraphModule

classmethod _factory(size_in: int, size_out: int, name: Optional[str] = None, computational_module: Optional[Any] = None, *args, **kwargs) → rockpool.graph.graph_base.GraphModuleBase

Build a new GraphModule or GraphModule subclass, with new input and output GraphNode s created automatically

Use this factory method to construct a new GraphModule from scratch, which needs new input and output GraphNode s created automatically. This helper method will be inherited by new GraphModule subclasses, and will act as factory methods also for your custom GraphModule subclass.

Parameters

• size_in (int) – The number of input GraphNode s to create and attach

• size_out (int) – The number of output GraphNode s to create and attach

• name (str, optional) – An arbitrary name to attach to this GraphModule, defaults to None

• computational_module (Optional[Module], optional) – A rockpool computational module that forms the “generator” of this graph module, defaults to None

Returns

The newly constructed GraphModule or GraphModule subclass

Return type

GraphModuleBase

add_input(node: rockpool.graph.graph_base.GraphNode) → None

Add a GraphNode as an input source to this module, and connect it

The new node will be appended after the last current input node. The node will be connected with this GraphModule as a sink.

Parameters

node (GraphNode) – The node to add as an input source

add_output(node: rockpool.graph.graph_base.GraphNode) → None

Add a GraphNode as an output of this module, and connect it

The new node will be appended after the last current output node. The node will be connected with this GraphModule as a source.

Parameters

node (GraphNode) – The node to add as an output

biases: Optional[Union[float, numpy.ndarray, torch.Tensor]] = None

The biases (Nout,) encapsulaed by this module

Type

FloatVector

clear_inputs() → None

Remove all GraphNode s as inputs of this module

clear_outputs() → None

Remove all GraphNode s as outputs of this module

computational_module: Module

The computational module that acts as the source for this graph module

Type

Module

input_nodes: SetList['GraphNode']

The input nodes attached to this module

Type

SetList[GraphNode]

name: str

An arbitrary name attached to this specific GraphModule

Type

str

output_nodes: SetList['GraphNode']

The output nodes attached to this module

Type

SetList[GraphNode]

remove_input(node: rockpool.graph.graph_base.GraphNode) → None

Remove a GraphNode as an input of this module, and disconnect it

The node will be disconnected from this GraphModule as a sink, and will be removed from the module.

Parameters

node (GraphNode) – The node to remove. If this node exists as an input to the module, it will be removed.

remove_output(node: rockpool.graph.graph_base.GraphNode) → None

Remove a GraphNode as an output of this module, and disconnect it

The node will be disconnected from this GraphModule as a source, and will be removed from the module.

Parameters

node (GraphNode) – The node to remove. If this node exists as an output to the module, it will be removed.

weights: Union[float, numpy.ndarray, torch.Tensor]

The linear weights (Nin, Nout) encapsulated by this module

Type

FloatVector