feat: Exa.TrkX abstract pipeline & metric hook

[benjaminhuth]

This Does the following:

• Abstracts the Exa.TrkX pipeline from the examples algorithm to the plugin. Exposes the pipeline to python, so it is possible to run and test the pipeline without the examples framework (e.g., directly with *.pyg or *.csv files):

spacepoints = list(np.random.uniform((100,3)).flatten())
spacepointIDs = list(np.arange(100))

pipeline = acts.examples.Pipeline(emb, [flt,gnn], trk, acts.logging.VERBOSE)
pipeline.run(spacepoints, spacepointIDs)

• Adds a interface for a pipeline hook, that is invoked after each stage with nodes and edges as arguments.
• Adds a implementation of the hook that prints the edge-based metrics of the graph with respect to the truth graph
• Extends the examples algorithm so it can generate the truth graph and a metric-hook:

08:52:52    EdgeClassifi   VERBOSE   Memory (used / total) [in MB]: 22155.4 / 40377.2
08:52:52    MetricsHook    INFO      Metrics for total graph:
08:52:52    MetricsHook    INFO      Efficiency=0.579324, purity=0.319214
08:52:52    MetricsHook    INFO      Metrics for target graph (pT > 0.5 GeV, nHits >= 3):
08:52:52    MetricsHook    INFO      Efficiency=0.968275, purity=0.138659
08:52:52    EdgeClassifi   DEBUG     Start edge classification

[codecov]

Codecov Report

Merging #2392 (6e11e17) into main (ed0bb27) will decrease coverage by 0.02%.
Report is 1 commits behind head on main.
The diff coverage is 63.15%.

@@            Coverage Diff             @@
##             main    #2392      +/-   ##
==========================================
- Coverage   49.69%   49.68%   -0.02%     
==========================================
  Files         454      454              
  Lines       25796    25799       +3     
  Branches    11850    11849       -1     
==========================================
- Hits        12820    12818       -2     
- Misses       4575     4580       +5     
  Partials     8401     8401              

Files Changed	Coverage Δ
...nclude/Acts/Vertexing/AdaptiveGridTrackDensity.hpp	100.00% <ø> (ø)
...nclude/Acts/Vertexing/GaussianGridTrackDensity.hpp	100.00% <ø> (ø)
...nclude/Acts/Vertexing/AdaptiveGridTrackDensity.ipp	61.15% <56.81%> (-1.26%)	⬇️
...nclude/Acts/Vertexing/GaussianGridTrackDensity.ipp	66.40% <84.61%> (-1.54%)	⬇️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[github-actions]

📊: Physics performance monitoring for 6e11e17

Full contents

physmon summary

• ✅ CKF truth_smeared
• ✅ IVF truth_smeared
• ✅ AMVF truth_smeared
• ✅ Track Summary CKF truth_smeared
• ✅ Seeding truth_estimated
• ✅ CKF truth_estimated
• ✅ IVF truth_estimated
• ✅ AMVF truth_estimated
• ✅ Track Summary CKF truth_estimated
• ✅ Seeding seeded
• ✅ CKF seeded
• ✅ IVF seeded
• ✅ AMVF seeded
• ✅ AMVF (+grid seeder) seeded
• ✅ Track Summary CKF seeded
• ✅ Seeding orthogonal
• ✅ CKF orthogonal
• ✅ IVF orthogonal
• ✅ AMVF orthogonal
• ✅ Track Summary CKF orthogonal
• ✅ Ambisolver seeded
• ✅ Ambisolver orthogonal
• ✅ Seeding ttbar
• ✅ CKF ttbar
• ✅ Ambisolver
• ✅ AMVF ttbar
• ✅ AMVF (+grid seeder) ttbar
• ✅ Truth tracking (GSF)
• ✅ Truth tracking
• ✅ Particles inital fatras
• ✅ Particles final fatras
• ✅ Particles inital geant4
• ✅ Particles final geant4

[benjaminhuth]

@xju2 This implements part of the things we discussed some time ago at CERN.

[paulgessinger]

Here's a few comments on the naming and the interface specification for these pipeline hooks.

Just to make sure the idea: the non-examples-FW python bindings are still delivered via the examples build. But the idea is that you can launch this without running the examples sequence by calling this pipeline infrastructure directly?

[benjaminhuth]

Just to make sure the idea: the non-examples-FW python bindings are still delivered via the examples build. But the idea is that you can launch this without running the examples sequence by calling this pipeline infrastructure directly?

Yes, since we don't have pure core bindings, I think thats the only possible way, right? This allows testing the pipeline and the hook directly with e.g. the training data, which is quite useful for me.