GNN Plugin

Detailed Description

Classes
class	ActsPlugins::BoostTrackBuilding
	Track building implementation using Boost. More...
class	ActsPlugins::CudaTrackBuilding
	Track building implementation using CUDA. More...
struct	ActsPlugins::Device
	A simple device description struct. More...
class	ActsPlugins::EdgeClassificationBase
	Base class for edge classification stage in GNN pipeline. More...
struct	ActsPlugins::ExecutionContext
	Capture the context of the execution. More...
class	ActsPlugins::GnnHook
	Hook interface for GNN pipeline callbacks. More...
class	ActsPlugins::GnnPipeline
	Graph Neural Network pipeline for track finding. More...
struct	ActsPlugins::GnnTiming
	Timing information for GNN pipeline execution. More...
class	ActsPlugins::GraphConstructionBase
	Base class for graph construction algorithms. More...
class	ActsPlugins::GraphStoreHook
	Hook for storing graph data during GNN pipeline execution. More...
class	ActsPlugins::ModuleMapCuda
	CUDA-based module map for graph construction. More...
struct	ActsPlugins::NoEdgesError
	Error that is thrown if no edges are found. More...
class	ActsPlugins::OnnxEdgeClassifier
	ONNX-based edge classifier for GNN pipeline. More...
struct	ActsPlugins::PipelineTensors
	Struct that ties together the tensors used in the GNN pipeline. More...
class	ActsPlugins::Tensor< T >
	This is a very small, limited class that models a 2D tensor of arbitrary type. More...
class	ActsPlugins::TensorRTEdgeClassifier
	Edge classifier using TensorRT inference. More...
class	ActsPlugins::TorchEdgeClassifier
	Edge classifier using PyTorch inference. More...
class	ActsPlugins::TorchMetricLearning
	Graph construction using PyTorch metric learning. More...
class	ActsPlugins::TrackBuildingBase
	Base class for track building implementations. More...
class	ActsPlugins::TruthGraphMetricsHook
	Hook for calculating metrics using truth graph information. More...

Functions
std::pair< Tensor< std::int64_t >, std::optional< Tensor< float > > >	ActsPlugins::applyEdgeLimit (const Tensor< std::int64_t > &edgeIndex, const std::optional< Tensor< float > > &edgeFeatures, std::size_t maxEdges, std::optional< cudaStream_t > stream)
	Apply a limit on the number of edges consistently on edgeIndex and edgeFeatures.
std::pair< Tensor< float >, Tensor< std::int64_t > >	ActsPlugins::applyScoreCut (const Tensor< float > &scores, const Tensor< std::int64_t > &edgeIndex, float cut, std::optional< cudaStream_t > stream={})
	Apply a score cut to the tensor and return a new tensor with the values that satisfy the cut.
std::ostream &	ActsPlugins::operator<< (std::ostream &os, Device device)
	Stream operator for Device.
void	ActsPlugins::sigmoid (Tensor< float > &tensor, std::optional< cudaStream_t > stream={})
	Element-wise sigmoid function for float cpu tensors.

Function Documentation

applyEdgeLimit()

std::pair< Tensor< std::int64_t >, std::optional< Tensor< float > > > ActsPlugins::applyEdgeLimit	(	const Tensor< std::int64_t > &	edgeIndex,
		const std::optional< Tensor< float > > &	edgeFeatures,
		std::size_t	maxEdges,
		std::optional< cudaStream_t >	stream )

Apply a limit on the number of edges consistently on edgeIndex and edgeFeatures.

Parameters

edgeIndex	The edge index tensor
edgeFeatures	The edge feature tensor
maxEdges	The edge limit to apply
stream	The stream to use for operation in case of CUDA

Returns

Pair of limited edge index tensor and optional limited edge features tensor

applyScoreCut()

std::pair< Tensor< float >, Tensor< std::int64_t > > ActsPlugins::applyScoreCut	(	const Tensor< float > &	scores,
		const Tensor< std::int64_t > &	edgeIndex,
		float	cut,
		std::optional< cudaStream_t >	stream = {} )

Apply a score cut to the tensor and return a new tensor with the values that satisfy the cut.

Parameters

scores	The edge score tensor
edgeIndex	The edge index tensor
cut	The score cut value which edges to accept
stream	The stream to use for the operation in case of CUDA

Returns

Pair of filtered score and edge index tensors containing only edges above threshold

operator<<()

std::ostream & ActsPlugins::operator<<	(	std::ostream &	os,
		Device	device )

Stream operator for Device.

Parameters

os	Output stream
device	Device to output

Returns

Reference to output stream

sigmoid()

void ActsPlugins::sigmoid	(	Tensor< float > &	tensor,
		std::optional< cudaStream_t >	stream = {} )

Element-wise sigmoid function for float cpu tensors.

Parameters

tensor	The tensor to apply the sigmoid function to
stream	The stream to use for the operation in case of CUDA