fix: Spacepoint-Track association in GNN algorithm

Examples/Algorithms/TrackFindingGnn/src/TrackFindingAlgorithmGnn.cpp

@@ -226,7 +226,7 @@ ProcessCode TrackFindingAlgorithmGnn::execute(
     onetrack.reserve(candidate.size());
 
     for (auto i : candidate) {
-      for (const auto& sl : spacePoints.at(i).sourceLinks()) {
+      for (const auto& sl : sortedSpacePoints.at(i).sourceLinks()) {
         onetrack.push_back(sl.template get<IndexSourceLink>().index());
       }
     }

Examples/Scripts/Python/gnn.py

@@ -7,6 +7,7 @@
 import acts
 import acts.examples
 import acts.examples.gnn
+from acts.examples.simulation import addDigiParticleSelection, ParticleSelectorConfig
 from acts.examples.reconstruction import addGnn, addSpacePointsMaking
 from acts.examples.gnn import (
     TorchMetricLearning,
@@ -43,6 +44,16 @@ def runGnnMetricLearning(
         s=s,
     )
 
+    addDigiParticleSelection(
+        s,
+        ParticleSelectorConfig(
+            pt=(1.0 * u.GeV, None),
+            eta=(-3.0, 3.0),
+            measurements=(7, None),
+            removeNeutral=True,
+        ),
+    )
+
     addSpacePointsMaking(
         s,
         geoSelectionConfigFile=geometrySelection,

Examples/Scripts/Python/gnn4itk_example.py

@@ -16,6 +16,7 @@
 import acts
 import acts.examples
 from acts.examples.reconstruction import addGnn
+from acts.examples.simulation import ParticleSelectorConfig, addDigiParticleSelection
 from acts.examples.gnn import (
     ModuleMapCuda,
     CudaTrackBuilding,
@@ -81,6 +82,15 @@ def runGNN4ITk(
         )
     )
 
+    # Select primary particles with minimum 7 hits and 1 GeV pT for efficiency evaluation
+    s.addWhiteboardAlias("particles_simulated_selected", "particles")
+    particleSelectorConfig = ParticleSelectorConfig(
+        pt=(1.0 * u.GeV, None),
+        hits=(7, None),
+        removeSecondaries=True,
+    )
+    addDigiParticleSelection(s, particleSelectorConfig, logLevel=logLevel)
+
     # Configure GNN stages for module map workflow
     # All parameters hardcoded based on ITk configuration
 

Examples/Scripts/Python/gnn_module_map_odd.py

@@ -25,7 +25,7 @@
     ParticleConfig,
     addFatras,
     addDigitization,
-    addGenParticleSelection,
+    addDigiParticleSelection,
     ParticleSelectorConfig,
 )
 from acts.examples.reconstruction import addGnn, addSpacePointsMaking
@@ -94,16 +94,6 @@ def runGnnModuleMap(
         rnd=rnd,
     )
 
-    addGenParticleSelection(
-        s,
-        ParticleSelectorConfig(
-            rho=(0.0, 24 * u.mm),
-            absZ=(0.0, 1.0 * u.m),
-            eta=(-3.0, 3.0),
-            pt=(150 * u.MeV, None),
-        ),
-    )
-
     # FATRAS simulation
     addFatras(
         s,
@@ -127,6 +117,16 @@ def runGnnModuleMap(
         logLevel=acts.logging.INFO,
     )
 
+    addDigiParticleSelection(
+        s,
+        ParticleSelectorConfig(
+            pt=(1.0 * u.GeV, None),
+            eta=(-3.0, 3.0),
+            measurements=(7, None),
+            removeNeutral=True,
+        ),
+    )
+
     addSpacePointsMaking(
         s,
         trackingGeometry,

Python/Examples/python/reconstruction.py

@@ -1819,6 +1819,7 @@ def addTrackWriters(
     writeStates: bool = False,
     writeFitterPerformance: bool = False,
     writeFinderPerformance: bool = False,
+    writeFinderNTuple: bool = False,
     logLevel: Optional[acts.logging.Level] = None,
     writeCovMat=False,
 ):
@@ -1879,6 +1880,17 @@ def addTrackWriters(
             )
             s.addWriter(trackFinderPerfWriter)
 
+        if writeFinderNTuple:
+            nTupleWriter = RootTrackFinderNTupleWriter(
+                level=customLogLevel(),
+                inputTracks="tracks",
+                inputParticles="particles_selected",
+                inputParticleMeasurementsMap="particle_measurements_map",
+                inputTrackParticleMatching="track_particle_matching",
+                filePath=str(Path(outputDirRoot) / f"ntuple_finding_{name}.root"),
+            )
+            s.addWriter(nTupleWriter)
+
     if outputDirCsv is not None:
         outputDirCsv = Path(outputDirCsv)
         if not outputDirCsv.exists():
@@ -1999,24 +2011,24 @@ def addGnn(
     s.addWhiteboardAlias("protoTracks", findingAlg.config.outputProtoTracks)
 
     # Convert proto tracks to tracks
-    s.addAlgorithm(
-        acts.examples.ProtoTracksToTracks(
-            level=customLogLevel(),
-            inputProtoTracks="protoTracks",
-            inputMeasurements="measurements",
-            outputTracks="tracks",
-        )
+    convAlg = acts.examples.ProtoTracksToTracks(
+        level=customLogLevel(),
+        inputProtoTracks=findingAlg.config.outputProtoTracks,
+        inputMeasurements="measurements",
+        outputTracks="gnn-tracks",
     )
+    s.addAlgorithm(convAlg)
+    s.addWhiteboardAlias("tracks", convAlg.config.outputTracks)
 
     # Truth matching
     matchAlg = acts.examples.TrackTruthMatcher(
         level=customLogLevel(),
-        inputTracks="tracks",
-        inputParticles="particles",
+        inputTracks=convAlg.config.outputTracks,
+        inputParticles="particles_selected",
         inputMeasurementParticlesMap="measurement_particles_map",
         outputTrackParticleMatching="gnn_track_particle_matching",
         outputParticleTrackMatching="gnn_particle_track_matching",
-        doubleMatching=True,
+        doubleMatching=False,
     )
     s.addAlgorithm(matchAlg)
     s.addWhiteboardAlias(
@@ -2026,21 +2038,16 @@ def addGnn(
         "particle_track_matching", matchAlg.config.outputParticleTrackMatching
     )
 
-    # Optional performance writer
-    if outputDirRoot is not None:
-        assert (
-            ACTS_EXAMPLES_ROOT_AVAILABLE
-        ), "ROOT output requested but ROOT is not available"
-        s.addWriter(
-            RootTrackFinderNTupleWriter(
-                level=customLogLevel(),
-                inputTracks="tracks",
-                inputParticles="particles",
-                inputParticleMeasurementsMap="particle_measurements_map",
-                inputTrackParticleMatching=matchAlg.config.outputTrackParticleMatching,
-                filePath=str(Path(outputDirRoot) / "performance_track_finding.root"),
-            )
-        )
+    addTrackWriters(
+        s,
+        name="gnn",
+        tracks=convAlg.config.outputTracks,
+        outputDirRoot=outputDirRoot,
+        outputDirCsv=None,
+        writeFinderPerformance=True,
+        writeSummary=False,
+        writeFinderNTuple=True,
+    )
 
     return s
 

Python/Examples/tests/root_file_hashes.txt

@@ -74,9 +74,12 @@ test_root_material_writer__material.root: e3b0c44298fc1c149afbf4c8996fb92427ae41
 test_root_clusters_writer[configPosConstructor]__clusters.root: e842df4fe04eefff3df5f32cd1026e93286be62b8040dc700a2aff557c56dec8
 test_root_clusters_writer[configKwConstructor]__clusters.root: e842df4fe04eefff3df5f32cd1026e93286be62b8040dc700a2aff557c56dec8
 test_root_clusters_writer[kwargsConstructor]__clusters.root: e842df4fe04eefff3df5f32cd1026e93286be62b8040dc700a2aff557c56dec8
-test_gnn_metric_learning[gpu]__performance_track_finding.root: 3f0fb36af55441994a154ea2a93978ba1930d4e87bf043f8ae9527e283bf1894
-test_gnn_module_map[gpu-torch]__performance_track_finding.root: db5c884283f5700cc2308bdf722e35b203ae018cbf10950fb71483c253685a3a
-test_gnn_module_map[gpu-onnx]__performance_track_finding.root: 80e6651bb1249407d675ba2f7fd4f2af01c4fcb9c393e7f4e526212a0102b545
+test_gnn_metric_learning[gpu]__performance_finding_gnn.root: c17fb877bb165e28db0a2b99881763093e7fc9a707c045feb6a6a6b68e0dd660
+test_gnn_metric_learning[gpu]__ntuple_finding_gnn.root: 3f0fb36af55441994a154ea2a93978ba1930d4e87bf043f8ae9527e283bf1894
+test_gnn_module_map[gpu-torch]__ntuple_finding_gnn.root: ae0c828b57f4d7b7e608fa92175af418647aece75b7ac1946bf30eb8e617d046
+test_gnn_module_map[gpu-torch]__performance_finding_gnn.root: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+test_gnn_module_map[gpu-onnx]__ntuple_finding_gnn.root: 664be95535722685effd783b148ed2cec61906aee20f5a531bedbd5253d2b101
+test_gnn_module_map[gpu-onnx]__performance_finding_gnn.root: e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
 test_ML_Ambiguity_Solver__performance_finding_ambiML.root: c17fb877bb165e28db0a2b99881763093e7fc9a707c045feb6a6a6b68e0dd660
 test_truth_tracking_kalman[generic-0.0]__trackstates_kf.root: 42a49abcef0277ca061350cc03c1ac9841e119055e2778662ec6a731e316ecd2
 test_truth_tracking_kalman[generic-0.0]__tracksummary_kf.root: bde974e07d033f9bb8e078c180d288c996f69590d6b4516625e1ec0276f327a7

Python/Examples/tests/test_examples.py

@@ -1095,8 +1095,9 @@ def test_gnn_metric_learning(tmp_path, trk_geo, field, assert_root_hash, hardwar
     if hardware == "cpu":
         pytest.skip("CPU not yet supported")
 
-    root_file = "performance_track_finding.root"
-    assert not (tmp_path / root_file).exists()
+    root_files = ["performance_finding_gnn.root", "ntuple_finding_gnn.root"]
+    for f in root_files:
+        assert not (tmp_path / f).exists()
 
     # Check if models exist using MODEL_STORAGE environment variable
     model_storage = os.environ.get("MODEL_STORAGE")
@@ -1133,10 +1134,11 @@ def test_gnn_metric_learning(tmp_path, trk_geo, field, assert_root_hash, hardwar
             print(e.output.decode("utf-8"))
         raise
 
-    rfp = tmp_path / root_file
-    assert rfp.exists()
+    for f in root_files:
+        rfp = tmp_path / f
+        assert rfp.exists()
 
-    assert_root_hash(root_file, rfp)
+        assert_root_hash(f, rfp)
 
 
 @pytest.mark.odd
@@ -1194,9 +1196,13 @@ def test_gnn_module_map(tmp_path, assert_root_hash, backend, hardware):
         )
 
     # Verify output
-    output_file = tmp_path / "performance_track_finding.root"
+    output_file = tmp_path / "performance_finding_gnn.root"
+    assert output_file.exists()
+    assert_root_hash("performance_finding_gnn.root", output_file)
+
+    output_file = tmp_path / "ntuple_finding_gnn.root"
     assert output_file.exists()
-    assert_root_hash("performance_track_finding.root", output_file)
+    assert_root_hash("ntuple_finding_gnn.root", output_file)
 
 
 @pytest.mark.odd