Functions
	MuonLearningOnnxRuntimeSvcCfg (flags, name="OnnxRuntimeSvc", **kwargs)
	GraphInferenceAlgCfg (flags, name="GraphInferenceAlg", **kwargs)
	GraphSPFilterToolCfg (flags, name="GraphSPFilterTool", **kwargs)
	GraphBucketFilterToolCfg (flags, name="GraphBucketFilterTool", **kwargs)
	SegmentEdgeClassifierToolCfg (flags, name="SegmentEdgeClassifierTool", **kwargs)
	SegmentEdgeInferenceAlgCfg (flags, name="SegmentEdgeInferenceAlg", **kwargs)
	DisplacedVertexInferenceToolCfg (flags, name="DisplacedVertexInferenceTool", **kwargs)
	DisplacedVertexCaloTowerCfg (flags)
	DisplacedVertexInferenceAlgCfg (flags, name="DisplacedVertexInferenceAlg", **kwargs)

Variables
str	DEFAULT_BUCKET_MODEL_PATH = "dev/MuonRecRTT/edgecnn_mu200.onnx"
float	DEFAULT_BUCKET_SCORE_THRESHOLD = 0.160
str	DEFAULT_BUCKET_SINGLE_OUTPUT_MODE = "logit"
str	DEFALT_DV_MODEL_PATH = "/cvmfs/atlas-nightlies.cern.ch/repo/data/data-art/MuonRecRTT/MlPhaseII/edge_gatv2.onnx"

Function Documentation

◆ DisplacedVertexCaloTowerCfg()

DisplacedVertexCaloTowerCfg ( flags )

Configure the calorimeter reconstruction used by the DV training converter.

Definition at line 168 of file InferenceConfig.py.

def DisplacedVertexCaloTowerCfg(flags):
    """Configure the calorimeter reconstruction used by the DV training converter.
    """
    result = ComponentAccumulator()
 
    # Same reconstruction chain used by MuonBucketDumpConfig.CaloCellsDumperCfg.
    from CaloRec.CaloRecoConfig import CaloRecoCfg
    result.merge(CaloRecoCfg(flags))
 
    from CaloRec.CaloTowerMakerConfig import CaloTowerMakerCfg
    result.getPrimaryAndMerge(CaloTowerMakerCfg(flags))
 
    return result
 
 

◆ DisplacedVertexInferenceAlgCfg()

DisplacedVertexInferenceAlgCfg	(		flags,
			name = "DisplacedVertexInferenceAlg",
		**	kwargs )

Configure a runnable event-level DisplacedVertex inference algorithm.

Definition at line 183 of file InferenceConfig.py.

def DisplacedVertexInferenceAlgCfg(flags, name="DisplacedVertexInferenceAlg", **kwargs):
    """Configure a runnable event-level DisplacedVertex inference algorithm."""
    result = ComponentAccumulator()
    do_calo_tower_build = kwargs.pop("DoCaloTowerBuild", True)
    do_ml_bucket_filter = kwargs.pop("DoMLBucketFilter", True)
    bucket_model_path = kwargs.pop("BucketModelPath", None)
    bucket_threshold = kwargs.pop("BucketThreshold", None)
    filtered_bucket_key = kwargs.pop("FilteredBucketKey", "FilteredMlBuckets")
    use_filtered_buckets_for_dv_graph = kwargs.pop("UseFilteredBucketsForDVGraph", False)
    alg_output_level = kwargs.get("OutputLevel", None)
    tool_kwargs = {}
    for key in (
        "ModelPath",
        "InputNodeName",
        "InputEdgeIndexName",
        "InputEdgeAttrName",
        "InputNMuonNodesName",
        "OutputName",
        "SingleOutputMode",
        "SegmentKey",
        "SpacePointKeys",
        "UseBucketSegmentSelection",
        "TowerContainerKey",
        "MinTowerEnergyMeV",
        "MaxTowerSegmentDR",
        "CaloRMaxMm",
        "CaloZMaxMm",
        "SectorModulo",
        "RequireEdges",
        "MaxEdges",
        "FallbackToAllSegments",
        "DebugDumpFirstNNodes",
        "DebugDumpFirstNEdges",
        "OutputLevel",
    ):
        if key in kwargs:
            tool_kwargs[key] = kwargs.pop(key)
 
    if isinstance(kwargs.get("InferenceTool"), dict):
        tool_kwargs.update(kwargs.pop("InferenceTool"))
        
    tower_key = tool_kwargs.get("TowerContainerKey", "CombinedTower")
    if do_calo_tower_build and tower_key:
        result.merge(DisplacedVertexCaloTowerCfg(flags))
 
    if do_ml_bucket_filter:
        bucket_filter_kwargs = {
            "WriteSpacePointKey": filtered_bucket_key,
            "ReadSpacePoints": "MuonSpacePoints",
        }
        if bucket_model_path is not None:
            bucket_filter_kwargs["ModelPath"] = bucket_model_path
        if bucket_threshold is not None:
            bucket_filter_kwargs["ScoreThreshold"] = bucket_threshold
        bucket_tool = result.popToolsAndMerge(
            GraphBucketFilterToolCfg(flags, **bucket_filter_kwargs)
        )
        result.merge(
            GraphInferenceAlgCfg(
                flags,
                name="DVBucketPrefilterAlg",
                InferenceTools=[bucket_tool],
            )
        )
        if use_filtered_buckets_for_dv_graph:
            tool_kwargs.setdefault("SpacePointKeys", [filtered_bucket_key])
        tool_kwargs.setdefault("UseBucketSegmentSelection", True)
 
    if "InferenceTool" not in kwargs:
        kwargs["InferenceTool"] = result.popToolsAndMerge(
            DisplacedVertexInferenceToolCfg(flags, **tool_kwargs)
        )
 
    if alg_output_level is not None:
        kwargs["OutputLevel"] = alg_output_level
 
    alg = CompFactory.MuonML.DVInferenceAlg(name=name, **kwargs)
    result.addEventAlgo(alg, primary=True)
    return result

◆ DisplacedVertexInferenceToolCfg()

DisplacedVertexInferenceToolCfg	(		flags,
			name = "DisplacedVertexInferenceTool",
		**	kwargs )

Configure the DisplacedVertex graph-level ONNX inference tool.
The current DV ONNX export consumes raw graph tensors with the contract
  x [N,7], edge_index [2,E], edge_attr [E,5], n_muon_nodes [1] -> logits [1]

Definition at line 129 of file InferenceConfig.py.

def DisplacedVertexInferenceToolCfg(flags, name="DisplacedVertexInferenceTool", **kwargs):
    """Configure the DisplacedVertex graph-level ONNX inference tool.
    The current DV ONNX export consumes raw graph tensors with the contract
      x [N,7], edge_index [2,E], edge_attr [E,5], n_muon_nodes [1] -> logits [1]
    """
    from AthOnnxComps.OnnxRuntimeSessionConfig import OnnxRuntimeSessionToolCfg
 
    result = ComponentAccumulator()
    model_path = kwargs.pop("ModelPath", DEFALT_DV_MODEL_PATH)
 
    result.merge(MuonLearningOnnxRuntimeSvcCfg(flags))
    kwargs.setdefault("ModelSession", result.popToolsAndMerge(
        OnnxRuntimeSessionToolCfg(flags, model_fname=model_path,
                                  OnnxRuntimeSvc=result.getService("OnnxRuntimeSvc"))))
    kwargs.setdefault("InputNodeName", "x")
    kwargs.setdefault("InputEdgeIndexName", "edge_index")
    kwargs.setdefault("InputEdgeAttrName", "edge_attr")
    kwargs.setdefault("InputNMuonNodesName", "n_muon_nodes")
    kwargs.setdefault("OutputName", "logits")
    kwargs.setdefault("SingleOutputMode", "logit")
    if "SpacePointKeys" not in kwargs:
        sp_containers = []
        if flags.Detector.GeometryMDT or flags.Detector.GeometryRPC or flags.Detector.GeometryTGC:
            sp_containers.append("MuonSpacePoints")
        elif flags.Detector.GeometryMM or flags.Detector.GeometrysTGC:
            sp_containers.append("NswSpacePoints")
        kwargs.setdefault("SpacePointKeys", sp_containers)
    kwargs.setdefault("UseBucketSegmentSelection", True)
    kwargs.setdefault("MinTowerEnergyMeV", 1000.0)
    kwargs.setdefault("MaxTowerSegmentDR", 0.4)
    kwargs.setdefault("CaloRMaxMm", 4250.0)
    kwargs.setdefault("CaloZMaxMm", 6500.0)
    kwargs.setdefault("SectorModulo", 16)
    kwargs.setdefault("RequireEdges", False)
    tool = CompFactory.MuonML.DVInferenceToolBase(name, **kwargs)
    result.setPrivateTools(tool)
    return result
 
 

◆ GraphBucketFilterToolCfg()

GraphBucketFilterToolCfg	(		flags,
			name = "GraphBucketFilterTool",
		**	kwargs )

Definition at line 38 of file InferenceConfig.py.

def GraphBucketFilterToolCfg(flags, name ="GraphBucketFilterTool", **kwargs):
 
    from AthOnnxComps.OnnxRuntimeSessionConfig import OnnxRuntimeSessionToolCfg
 
    result = ComponentAccumulator()
    model_path = kwargs.pop("ModelPath", DEFAULT_BUCKET_MODEL_PATH)
    single_output_mode = kwargs.pop("SingleOutputMode", None)
    if single_output_mode is not None:
        if "SingleOutputIsLogit" in kwargs:
            raise ValueError(
                "Specify either SingleOutputMode or SingleOutputIsLogit, not both."
            )
        if single_output_mode not in ("logit", "prob"):
            raise ValueError(
                "SingleOutputMode must be 'logit' or 'prob', got "
                f"{single_output_mode!r}."
            )
        kwargs["SingleOutputIsLogit"] = (single_output_mode == "logit")
    
    result.merge(MuonLearningOnnxRuntimeSvcCfg(flags))
    kwargs.setdefault("ModelSession", result.popToolsAndMerge(
        OnnxRuntimeSessionToolCfg(flags, model_fname=model_path,
                                  OnnxRuntimeSvc=result.getService("OnnxRuntimeSvc"))))
    kwargs.setdefault("OutputLevel", 3)  # INFO level (1=VERBOSE, 2=DEBUG, 3=INFO, 4=WARNING, 5=ERROR, 6=FATAL)
    kwargs.setdefault("ScoreThreshold", DEFAULT_BUCKET_SCORE_THRESHOLD)
    kwargs.setdefault("OutputName", "logits")
    kwargs.setdefault("SingleOutputIsLogit", DEFAULT_BUCKET_SINGLE_OUTPUT_MODE == "logit")
    
    the_tool = CompFactory.MuonML.GraphBucketFilterTool(name, **kwargs)
    result.setPrivateTools(the_tool)
    return result
 
 

◆ GraphInferenceAlgCfg()

GraphInferenceAlgCfg	(		flags,
			name = "GraphInferenceAlg",
		**	kwargs )

Definition at line 20 of file InferenceConfig.py.

def GraphInferenceAlgCfg(flags, name = "GraphInferenceAlg", **kwargs):
    result  = ComponentAccumulator()
    the_alg = CompFactory.MuonML.InferenceAlg(name, **kwargs)
    result.addEventAlgo(the_alg, primary = True)
    return result
 

◆ GraphSPFilterToolCfg()

GraphSPFilterToolCfg	(		flags,
			name = "GraphSPFilterTool",
		**	kwargs )

Definition at line 26 of file InferenceConfig.py.

def GraphSPFilterToolCfg(flags, name ="GraphSPFilterTool", **kwargs):
    
    from AthOnnxComps.OnnxRuntimeSessionConfig import OnnxRuntimeSessionToolCfg
 
    result = ComponentAccumulator()
    kwargs.setdefault("ModelSession", result.popToolsAndMerge(OnnxRuntimeSessionToolCfg(flags, model_fname="/eos/atlas/atlascerngroupdisk/data-art/grid-input/MuonRecRTT/TestModel.onnx")))
    kwargs.setdefault("MLFilterCut", -3.6) # Working point cut
    
    the_tool = CompFactory.MuonML.GraphSPFilterTool(name, **kwargs)
    result.setPrivateTools(the_tool)
    return result
 

◆ MuonLearningOnnxRuntimeSvcCfg()

MuonLearningOnnxRuntimeSvcCfg	(		flags,
			name = "OnnxRuntimeSvc",
		**	kwargs )

Configure the shared ONNX Runtime service used by MuonLearning tools.

Definition at line 12 of file InferenceConfig.py.

def MuonLearningOnnxRuntimeSvcCfg(flags, name="OnnxRuntimeSvc", **kwargs):
    """Configure the shared ONNX Runtime service used by MuonLearning tools."""
    result = ComponentAccumulator()
    kwargs.setdefault("LogLevel", 3)
    svc = CompFactory.AthOnnx.OnnxRuntimeSvc(name, **kwargs)
    result.addService(svc, primary=False, create=True)
    return result
 

◆ SegmentEdgeClassifierToolCfg()

SegmentEdgeClassifierToolCfg	(		flags,
			name = "SegmentEdgeClassifierTool",
		**	kwargs )

Definition at line 71 of file InferenceConfig.py.

def SegmentEdgeClassifierToolCfg(flags, name="SegmentEdgeClassifierTool", **kwargs):
    from AthOnnxComps.OnnxRuntimeSessionConfig import OnnxRuntimeSessionToolCfg
 
    result = ComponentAccumulator()
    model_path = kwargs.pop("ModelPath", "MuonInference/models/edge_gnn_refit_top01_from_t0020.onnx")
    result.merge(MuonLearningOnnxRuntimeSvcCfg(flags))
    kwargs.setdefault("ModelSession", result.popToolsAndMerge(
        OnnxRuntimeSessionToolCfg(flags, model_fname=model_path,
                                  OnnxRuntimeSvc=result.getService("OnnxRuntimeSvc"))))
    # Keep the same ONNX/model properties used by GraphBucketFilterToolCfg in this file.
    kwargs.setdefault("InputNodeName", "x")
    kwargs.setdefault("InputEdgeIndexName", "edge_index")
    kwargs.setdefault("InputEdgeAttrName", "edge_attr")
    kwargs.setdefault("OutputName", "logits")
    kwargs.setdefault("MaxDeltaThetaDeg", 35.0)
    kwargs.setdefault("MaxDeltaSector", 1)
    kwargs.setdefault("SectorModulo", 16)
    kwargs.setdefault("ReadSpacePoints", "MuonSpacePoints")
    # Zero leaves pre-inference pruning disabled.
    kwargs.setdefault("MaxSegmentsPerBucket", 0)
    kwargs.setdefault("MaxEdgesPerNodeBeforeInference", 0)
    kwargs.setdefault("MaxEdgesPerTargetChamberBeforeInference", 0)
    tool = CompFactory.MuonML.SegmentEdgeClassifierTool(name, **kwargs)
    result.setPrivateTools(tool)
    return result
 
 

◆ SegmentEdgeInferenceAlgCfg()

SegmentEdgeInferenceAlgCfg	(		flags,
			name = "SegmentEdgeInferenceAlg",
		**	kwargs )

Definition at line 98 of file InferenceConfig.py.

def SegmentEdgeInferenceAlgCfg(flags, name="SegmentEdgeInferenceAlg", **kwargs):
    result = ComponentAccumulator()
    # Accept a model-path shortcut or an EdgeClassifierTool settings
    edge_tool_kwargs = {}
    if "EdgeModelPath" in kwargs:
        edge_tool_kwargs["ModelPath"] = kwargs.pop("EdgeModelPath")
    # Unwrap a settings mapping supplied for EdgeClassifierTool.
    if isinstance(kwargs.get("EdgeClassifierTool"), dict):
        edge_tool_kwargs.update(kwargs.pop("EdgeClassifierTool"))
 
    if "EdgeClassifierTool" not in kwargs:
        kwargs["EdgeClassifierTool"] = result.popToolsAndMerge(
            SegmentEdgeClassifierToolCfg(flags, **edge_tool_kwargs))
    kwargs.setdefault("SegmentKey", "MsSegments")
    kwargs.setdefault("PairGateDecoration", "MsSegments.mlTrackComponent")
    # An empty key disables this optional non-owning view. When enabled it
    # contains only segments incident to a post-classifier selected edge.
    kwargs.setdefault("FilteredSegmentKey", "")
    kwargs.setdefault("PairGateThreshold", 0.975)
    # Build components from mutual top-K edge associations 
    kwargs.setdefault("UseDegreeCappedComponents", False)
    kwargs.setdefault("RequireMutualTopKEdges", True)
    kwargs.setdefault("RecoverOrphanNodes", True)
    kwargs.setdefault("SeedAnchorsPerComponent", 0)
    kwargs.setdefault("AnchorInnermostLayer", True)
    kwargs.setdefault("MinSegmentsPerComponent", 2)
    alg = CompFactory.MuonML.SegmentEdgeInferenceAlg(name=name, **kwargs)
    result.addEventAlgo(alg, primary=True)
    return result
 
 

Variable Documentation

◆ DEFALT_DV_MODEL_PATH

str InferenceConfig.DEFALT_DV_MODEL_PATH = "/cvmfs/atlas-nightlies.cern.ch/repo/data/data-art/MuonRecRTT/MlPhaseII/edge_gatv2.onnx"

Definition at line 10 of file InferenceConfig.py.

◆ DEFAULT_BUCKET_MODEL_PATH

str InferenceConfig.DEFAULT_BUCKET_MODEL_PATH = "dev/MuonRecRTT/edgecnn_mu200.onnx"

Definition at line 6 of file InferenceConfig.py.

◆ DEFAULT_BUCKET_SCORE_THRESHOLD

float InferenceConfig.DEFAULT_BUCKET_SCORE_THRESHOLD = 0.160

Definition at line 7 of file InferenceConfig.py.

◆ DEFAULT_BUCKET_SINGLE_OUTPUT_MODE

str InferenceConfig.DEFAULT_BUCKET_SINGLE_OUTPUT_MODE = "logit"

Definition at line 8 of file InferenceConfig.py.

Functions

Variables