Make track algo priority order dynamic #17771

makortel · 2017-03-03T17:56:54Z

Currently the "track algo priority order", that is used in track collection merging to dictate the order of iterations, is hardcoded. The order is the one of run2 tracking, but phase1 and phase2 tracking have the iterations in different orders. This PR proposes to make this iteration ordering dynamic via ESProduct TrackAlgoPriorityOrder that is configured via tracking eras using iterativeTkConfig.py. The TrackAlgoPriorityOrder starts with the hardcoded order, and reorders only the algos that are given as parameters.

Tested in CMSSW_9_1_X_2017-03-02-2300. There can be tiny changes in non-run2-tracking in cases the "best algo" gets changed by this reordering and the other track has slightly different parameters. Anything else (run2 tracking, HLT) should have no changes. (and all changes should be caused by the second commit alone, the first one being purely technical)

@rovere @VinInn

…oPriorityOrder ESProduct The order of (offline) tracking iterations depends on tracking era, so something like this is needed for proper bookkeeping of originalAlgo.

makortel · 2017-03-03T17:57:06Z

@cmsbuild, please test

cmsbuild · 2017-03-03T17:57:13Z

The tests are being triggered in jenkins.
https://cmssdt.cern.ch/jenkins/job/ib-any-integration/18133/console Started: 2017/03/03 18:58

cmsbuild · 2017-03-03T17:57:14Z

A new Pull Request was created by @makortel (Matti Kortelainen) for master.

It involves the following packages:

DQM/BeamMonitor
FastSimulation/Tracking
HLTrigger/Configuration
RecoHI/HiMuonAlgos
RecoHI/HiTracking
RecoLocalTracker/SubCollectionProducers
RecoMuon/Configuration
RecoMuon/L3MuonProducer
RecoTracker/FinalTrackSelectors
RecoTracker/IterativeTracking

@perrotta, @cmsbuild, @civanch, @lveldere, @silviodonato, @cvuosalo, @fwyzard, @ssekmen, @mdhildreth, @dmitrijus, @Martin-Grunewald, @slava77, @vanbesien, @davidlange6 can you please review it and eventually sign? Thanks.
@ghellwig, @felicepantaleo, @forthommel, @abbiendi, @MiheeJo, @Martin-Grunewald, @threus, @geoff-smith, @battibass, @jhgoh, @jlagram, @HuguesBrun, @OlivierBondu, @mandrenguyen, @rociovilar, @yduhm, @GiacomoSguazzoni, @rovere, @VinInn, @bellan, @jalimena, @mschrode, @dgulhan, @echapon, @trocino, @gbenelli, @jazzitup, @calderona, @nickmccoll, @yenjie, @kurtejung, @gpetruc, @matt-komm, @yetkinyilmaz, @bachtis this is something you requested to watch as well.
@Muzaffar, @davidlange6, @smuzaffar you are the release manager for this.

cms-bot commands are listed here #13028

cmsbuild · 2017-03-03T19:39:15Z

+1
Tested at: 156a2ef
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-17771/18133/summary.html

cmsbuild · 2017-03-03T19:39:18Z

Comparison job queued.

cmsbuild · 2017-03-03T20:25:28Z

Comparison is ready
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-17771/18133/summary.html

Martin-Grunewald · 2017-03-04T05:50:27Z

HLTrigger/Configuration/python/customizeHLTforCMSSW.py

 # CMSSW version specific customizations
 def customizeHLTforCMSSW(process, menuType="GRun"):
    # add call to action function in proper order: newest last!
    # process = customiseFor12718(process)
+    process = customiseForXXXXX(process)


Please replace, twice, XXXXX by 17771.

Whoops, fixed, thanks.

cmsbuild · 2017-03-04T17:43:57Z

Pull request #17771 was updated. @perrotta, @cmsbuild, @civanch, @lveldere, @silviodonato, @fwyzard, @ssekmen, @mdhildreth, @dmitrijus, @Martin-Grunewald, @slava77, @vanbesien, @davidlange6 can you please check and sign again.

makortel · 2017-03-04T17:44:17Z

@cmsbuild, please test

cmsbuild · 2017-03-04T17:44:31Z

The tests are being triggered in jenkins.
https://cmssdt.cern.ch/jenkins/job/ib-any-integration/18143/console Started: 2017/03/04 18:45

cmsbuild · 2017-03-04T19:24:34Z

+1
Tested at: 160bba7
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-17771/18143/summary.html

cmsbuild · 2017-03-04T19:24:36Z

Comparison job queued.

cmsbuild · 2017-03-04T21:05:50Z

Comparison is ready
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-17771/18143/summary.html

silviodonato · 2017-03-06T09:07:37Z

+1

dmitrijus · 2017-03-07T13:15:41Z

+1

perrotta · 2017-03-09T14:57:48Z

Tiny differences are observed in the phase 1 and phase 2 tracking distribution and none in the run2 ones, as expected.

Jenkins pretends there are also difference in HLT, see e.g.
https://cmssdt.cern.ch/SDT/jenkins-artifacts/baseLineComparisons/CMSSW_9_1_X_2017-03-03-1100+17771/18573/10824.0_TTbar_13+TTbar_13TeV_TuneCUETP8M1_2018_GenSimFull+DigiFull_2018+RecoFull_2018+ALCAFull_2018+HARVESTFull_2018/HLT__Tracking_ValidationWRTtp_hltIter0PFlowTrackSelectionHighPurity_hltAssociatorByHits_h_phipulleta_Mean.png
(where it is indeed hard to spot any difference by eyes in those histos, though...)
or
https://cmssdt.cern.ch/SDT/jenkins-artifacts/baseLineComparisons/CMSSW_9_1_X_2017-03-03-1100+17771/18573/21234.0_TTbar_14TeV+TTbar_14TeV_TuneCUETP8M1_2023D4_GenSimHLBeamSpotFull14+DigiFullTrigger_2023D4+RecoFullGlobal_2023D4+HARVESTFullGlobal_2023D4/Tracking__TrackParameters_highPurityTracks_pt_0to1_GeneralProperties_originalAlgorithm_GenTk.png
(where it looks like one track is actually selected by a different iteration)

@makortel : do you have any explanation for those tiny differences in HLT tracking (as you claimed they were not expected)?

slava77 · 2017-03-09T15:10:55Z

@makortel
do we expect further changes in the order of iterations in phase-1?

makortel · 2017-03-09T16:24:16Z

@slava77

do we expect further changes in the order of iterations in phase-1?

No(t for this year at least).

makortel · 2017-03-09T16:32:24Z

@perrotta The first plot (*_Mean, also *_Sigma) is produced by gaussian fits in the harvesting step. In practice if the input plots for one of them changes, it may induce tiny numerical changes in others that had no changes in their input histograms (IIRC caused by internal state of minuit not being reset between fit calls or something).

The second plot is from offline tracking, where tiny changes in the algo assignment are expected.

perrotta · 2017-03-10T11:35:21Z

+1
Because of #17771 (comment) and the answer in #17771 (comment) (indeed, the second plot was from offline tracking, not from HLT...)

makortel added 2 commits March 3, 2017 18:52

Replace trackAlgoPriorityOrder with a run-time configuration TrackAlg…

18e22ce

…oPriorityOrder ESProduct The order of (offline) tracking iterations depends on tracking era, so something like this is needed for proper bookkeeping of originalAlgo.

Set the priority order according to era

156a2ef

cmsbuild added this to the Next CMSSW_9_1_X milestone Mar 3, 2017

cmsbuild added comparison-pending dqm-pending fastsim-pending hlt-pending labels Mar 3, 2017

cmsbuild added orp-pending pending-signatures reconstruction-pending tests-pending labels Mar 3, 2017

cmsbuild added tests-started and removed tests-pending labels Mar 3, 2017

cmsbuild added tests-approved and removed tests-started labels Mar 3, 2017

cmsbuild added comparison-available and removed comparison-pending labels Mar 3, 2017

Martin-Grunewald reviewed Mar 4, 2017

View reviewed changes

Fix PR number

160bba7

cmsbuild added comparison-pending tests-pending and removed comparison-available tests-approved labels Mar 4, 2017

cmsbuild added tests-started and removed tests-pending labels Mar 4, 2017

cmsbuild added tests-approved and removed tests-started labels Mar 4, 2017

cmsbuild added comparison-available and removed comparison-pending labels Mar 4, 2017

cmsbuild added hlt-approved and removed hlt-pending labels Mar 6, 2017

cmsbuild added dqm-approved and removed dqm-pending labels Mar 7, 2017

cmsbuild added reconstruction-approved and removed reconstruction-pending labels Mar 10, 2017

davidlange6 merged commit 22f15e6 into cms-sw:master Mar 11, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Make track algo priority order dynamic #17771

Make track algo priority order dynamic #17771

makortel commented Mar 3, 2017

makortel commented Mar 3, 2017

cmsbuild commented Mar 3, 2017 •

edited

cmsbuild commented Mar 3, 2017

cmsbuild commented Mar 3, 2017

cmsbuild commented Mar 3, 2017

cmsbuild commented Mar 3, 2017

Martin-Grunewald Mar 4, 2017

makortel Mar 4, 2017

cmsbuild commented Mar 4, 2017

makortel commented Mar 4, 2017

cmsbuild commented Mar 4, 2017 •

edited

cmsbuild commented Mar 4, 2017

cmsbuild commented Mar 4, 2017

cmsbuild commented Mar 4, 2017

silviodonato commented Mar 6, 2017

dmitrijus commented Mar 7, 2017

perrotta commented Mar 9, 2017

slava77 commented Mar 9, 2017

makortel commented Mar 9, 2017

makortel commented Mar 9, 2017

perrotta commented Mar 10, 2017

Make track algo priority order dynamic #17771

Make track algo priority order dynamic #17771

Conversation

makortel commented Mar 3, 2017

makortel commented Mar 3, 2017

cmsbuild commented Mar 3, 2017 • edited

cmsbuild commented Mar 3, 2017

cmsbuild commented Mar 3, 2017

cmsbuild commented Mar 3, 2017

cmsbuild commented Mar 3, 2017

Martin-Grunewald Mar 4, 2017

Choose a reason for hiding this comment

makortel Mar 4, 2017

Choose a reason for hiding this comment

cmsbuild commented Mar 4, 2017

makortel commented Mar 4, 2017

cmsbuild commented Mar 4, 2017 • edited

cmsbuild commented Mar 4, 2017

cmsbuild commented Mar 4, 2017

cmsbuild commented Mar 4, 2017

silviodonato commented Mar 6, 2017

dmitrijus commented Mar 7, 2017

perrotta commented Mar 9, 2017

slava77 commented Mar 9, 2017

makortel commented Mar 9, 2017

makortel commented Mar 9, 2017

perrotta commented Mar 10, 2017

cmsbuild commented Mar 3, 2017 •

edited

cmsbuild commented Mar 4, 2017 •

edited