Topological Trigger Update 2024 April
Aims to update the Topological Trigger to 2024 conditions with no UT.
Following the discussion from the last WP3 presentation on the final NN configuration.
This is trained on 2024 HLt filtered samples and optimised to a combined rate of 30.4448 kHz. Individually, not accounting for overlaps, the rates are:
| line | rate |
|---|---|
| 2body Beauty Topo | 17.5589 |
| 3body Beauty Topo | 15.4210 |
| 2Body single muon Topo | 5.3856 |
| 3body Single muon Topo | 12.0681 |
Changes from current master:
-
Moved momentum cuts to the ParticleFilter for SpeedUp (Throughput Increase) -
Removed GhostProbability Cut due to global cut -
Removed TrackCHI2 cuts as per discussion in WP3 (Recommendation by WP2) -
Tightend DIRA cut from 0 to 0.9 (Throughput Increase) -
Relaxed MCorr cut from 10GeV to 12 GeV as per SL request (Analysis request) -
Removed LifeTime cut due to Software Maintainer request (For more stability in the qmt tests) -
Re-optimised Lipschitz constant and adjusted them in both neural networks -
General code clean up (Removed unused arguments i.e. make_pvs called in argument and function itself) -
Retrained on HLT1-filtered expected 2024 MC samples (no UT) -
Evaluated on HLT1-filtered expected 2024 MC samples (no UT) -
Truthmatching scheme for the training has been changed from 3 generations to 14 generations (searching for the same beauty hadron in all tracks up to 14 generations)
Efficiency Plots from the EfficiencyChecker corresponding to this working point: KstMuNu_CanRecoChildrenParentCut_TrueSim.pdf DstMuNu_CanRecoChildrenParentCut_TrueSim.pdfJPsiK_JPsi2MuMu_CanRecoChildrenParentCut_TrueSim.pdfJPsiPhi_CanRecoChildrenParentCut_TrueSim.pdf
goes together with lhcb-datapkg/ParamFiles!85