Hit disambiguation with SpacePointSolver Robert Sulej (FNAL/NCBJ)
Motivation: • FD hit disambiguation takes significant CPU time in ProtoDUNE events • It is also source of inefficiency for 2D clustering and 3D tracking high cosmic background: ~100k hits/event, ~70 long and crossing tracks • Try SpacePointSolver for matching objects between planes start with hits continue with higer level objects (make PMA faster) Implementation • SpacePointSolver running out of the box • Chris made all needed functionality during Physics Week • Disambiguation module: use hits associated to 3D points as a first reliable source of hit-wire segment association resolve remaining hits by associations made for hits in close neighborhood put leftovers on the first / copy to all allowed wire segments / or drop them
Space Point Solver „top”: XZ linecluster+PMA tracks, linecluster+PMA tracks, SpacePoints: red dots EM hits in black „ side ”: YZ, here is is th the real work rk 2D planes 3D orthogonal views • several points per collection hit for isochronous tracks, points fixed at wire crossings • not yet easy solution for everything, but very promissing for object associations
ProtoDUNE event full event beam particle
ProtoDUNE event based on SpacePointSolver FD disambiguation
ProtoDUNE: efficiency (only 80 events for today..) FD disambiguation SpacePointSolver based FD disambiguation SpacePointSolver based hit disambiguation cosmic tracks reconstruction (linecluster+PMA)
FD disambiguation =============================================================================================================================== TimeTracker printout (sec) Min Avg Max Median RMS nEvts =============================================================================================================================== Full event 371.099 485.6 671.889 485.951 89.7228 10 ------------------------------------------------------------------------------------------------------------------------------- source:RootInput(read) 0.00150542 0.00749291 0.013159 0.00716321 0.00362402 10 reco:rns:RandomNumberSaver 0.000124089 0.000325715 0.00111477 0.000186478 0.000298859 10 reco:caldata:DataPrepModule 6.15812 6.47112 7.39641 6.3202 0.361649 10 reco:gaushit:GausHitFinder 66.4812 83.4614 108.815 79.2166 12.9027 10 reco:hitfd:HitFinder35t 81.61 140.864 214.102 139.619 38.5684 10 reco:linecluster:LineCluster 1.60251 2.42541 4.14429 2.09614 0.796788 10 reco:emtrkmichelid:EmTrackMichelId 58.98 77.3793 126.822 73.8296 17.7905 10 reco:pmtrack:PMAlgTrackMaker 110.32 157.531 239.801 156.531 34.7789 10 reco:TriggerResults:TriggerResultInserter 6.748e-05 0.000118975 0.000204649 0.000109778 4.69442e-05 10 end_path:out1:RootOutput 8.684e-06 5.39718e-05 0.000126547 4.2725e-05 3.95037e-05 10 end_path:out1:RootOutput(write) 13.4487 17.4583 19.3223 18.384 1.85881 10 =============================================================================================================================== MemoryTracker summary (base-10 MB units used) Peak virtual memory usage (VmPeak) : 5026.18 MB Peak resident set size usage (VmHWM): 3443.03 MB Disambiguation using SpacePointSolver =============================================================================================================================== TimeTracker printout (sec) Min Avg Max Median RMS nEvts =============================================================================================================================== Full event 212.642 357.662 590.977 355.582 72.6383 40 ------------------------------------------------------------------------------------------------------------------------------- source:RootInput(read) 0.0034384 0.114365 0.2583 0.125503 0.0728791 40 reco:rns:RandomNumberSaver 0.000109423 0.000192507 0.000670115 0.000152965 0.000116677 40 reco:caldata:DataPrepModule 6.16531 6.56659 7.29999 6.53068 0.212751 40 reco:gaushit:GausHitFinder 42.6205 68.8812 103.278 68.7188 14.1731 40 reco:reco3d:SpacePointSolver 2.56435 6.25848 17.9495 5.33037 3.58018 40 reco:hitpdune:DisambigFromSpacePoints 1.79104 3.75687 6.93505 3.64876 1.2662 40 reco:linecluster:LineCluster 1.13328 2.52352 5.04337 2.50412 0.705175 40 reco:emtrkmichelid:EmTrackMichelId 57.1505 86.7008 143.323 81.129 18.5098 40 reco:pmtrack:PMAlgTrackMaker 90.1119 167.591 301.552 161.914 41.2546 40 reco:TriggerResults:TriggerResultInserter 5.0288e-05 0.000317919 0.0014757 0.000117304 0.000430709 40 end_path:out1:RootOutput 1.7469e-05 5.00584e-05 0.000333773 2.59585e-05 6.67579e-05 40 end_path:out1:RootOutput(write) 10.7917 15.2677 19.1317 15.2177 2.02879 40 =============================================================================================================================== MemoryTracker summary (base-10 MB units used) Peak virtual memory usage (VmPeak) : 6491.93 MB Peak resident set size usage (VmHWM): 4928.82 MB
DUNE FD: not that good with a simple approach based on SpacePointSolver FD disambiguation
DUNE FD: not that good with a simple approach
Summary • Christmass gift for ProtoDUNE: faster and better • Now in ProtoDUNE SP standard reco, will run it in MCC10 • Not sure why so poor in FD standard disambiguation tuned very well and not a slow part of the chain – good …but also no visible failures in eye-scanning of SpacePointSolver based hits (while efficiency histo says ~10% incorrect) • Next: implement clusters scorring in PMA • Seems that many other algorithms can use SpacePointSolver as a seed/hint. • Dealing with noisy/bad channels may be a good idea before ProtoDUNE data
Recommend
More recommend