Implementation of Fall17 cutbased Electron ID V2 #23477
Conversation
|
The code-checks are being triggered in jenkins. |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-23477/5025 |
|
A new Pull Request was created by @UAEDF-tomc (Tom Cornelis) for master. It involves the following packages: RecoEgamma/ElectronIdentification @perrotta, @cmsbuild, @slava77 can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
|
@cmsbuild please test @UAEDF-tomc |
|
The tests are being triggered in jenkins. |
|
Link to slides of latest update (30th of May 2018): Earlier updates preceding the final update: |
|
Comparison job queued. |
|
Comparison is ready Comparison Summary:
|
| @@ -0,0 +1,8 @@ | |||
| # |eta| min |eta| max effective area | |||
There was a problem hiding this comment.
please add some documentation comments similar to effAreaElectrons_cone03_pfNeuHadronsAndPhotons_92X
| const float _barrelC0, _endcapC0, _barrelCpt, _endcapCpt, _barrelCutOff; | ||
| bool _isRelativeIso; | ||
| EffectiveAreas _effectiveAreas; | ||
| edm::Handle<double> _rhoHandle; |
There was a problem hiding this comment.
please do not use leading underscore in variable names (trailing is good)
https://cms-cpt-software.web.cern.ch/cms-cpt-software/General/CodingAndStyleRules.pdf
There was a problem hiding this comment.
(note that in my original commit I had the leading underscores because I based my implementation on the already existing GsfEleEffAreaPFIsoCut.cc in the same directory, which also had leading underscores)
| #central_id_registry.register(cutBasedElectronID_Fall17_94X_V1_veto.idName, '43be9b381a8d9b0910b7f81a5ad8ff3a') | ||
| #central_id_registry.register(cutBasedElectronID_Fall17_94X_V1_loose.idName, '0b8456d622494441fe713a6858e0f7c1') | ||
| #central_id_registry.register(cutBasedElectronID_Fall17_94X_V1_medium.idName, 'a238ee70910de53d36866e89768500e9') | ||
| #central_id_registry.register(cutBasedElectronID_Fall17_94X_V1_tight.idName, '4acb2d2796efde7fba75380ce8823fc2') |
There was a problem hiding this comment.
IIUC, the comments above imply that these lines with MD5 hash values should be uncommented
|
@UAEDF-tomc |
Thank you for the links. |
|
I think I understand now: these _electron_id_modules_WorkingPoints.WorkingPoints define the ones which are used to built some integer which contain both veto, loose, medium and tight. Of course this is limited to 4 workingpoints as they have to fit in the integer, and it does not make sense to have V1 and V2 both listed here, so I only keep V2. The V1 should still be saved in the NanoAOD as it is included in the userIntsFromBool below. |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-23477/5248 |
|
@cmsbuild please test |
|
The tests are being triggered in jenkins. |
|
Comparison job queued. |
|
Comparison is ready Comparison Summary:
|
|
+1
|
|
merge |
Electron MVA V2 (as in cms-sw#23700 cms-sw#23746 cms-sw#23767) All event based variables are now handeled by the new helper class Hardcoding of variables and clips removed for Fall17 Generalized Fall17Iso and Fall17NoIso Configurable categories and start to move Spring16 MVA Dedicated variable files for Spring16 and Fall17 Removed Spring16 C++ code All year specific C++ code removed Small improvements Cleared some inconsistencies in variable definitions. All IDs now validated. Restructured VID python code a bit New MVAVariableManager helper class Forgot to actually add the class go git The raw BDT score is now stored as well Spring15 PhotonMVA uses now MVAVariableManager Almost there for photon MVA All Run2 Photon MVA IDs use now the new PhotonMVAEstimator class Add variable files Rewrote some python, ElectronID cuts on raw MWA values and autodetection of gzipped weight files Implemented V2 MV Ele ID Cut values are now found with parser as well Include ntupizer for training cleaned up a bit Training Ntuplizer finished Use auto generatetd config files Finalized ntuplizer Auto updated Training ntuplizer Updated training ntuplizer Fix in training ntuplizer fixed wp90 iso V1 parameters Fix Added MVA output Validation ntuplizer for photons Bugfix Improved exception handling in GBRForestToors Changed egammaObjectModificationsInMiniAOD_cff.py to comply with new config files Add two lines in miniAOD_tools.py to cope with ElectronMVAVariableHelper Fixed Fall17V1 variables Addressed comments following visual code inspection Simplified code Included Fall17 MVAs in MiniAOD and NanoAOD Added electronMVAVariableHelper to relevant processes Revisited casting of candidates to electrons/photons Fixup - prefer dynamic_cast if possible Corrected the second C style cast in the photons Avoid undefined category for photons Changed category not defined error to warning Changed mvaValue if category not found to -999 Fixed possible memory leak Drop MVAObjectCache in MVAValueMapProducer Commit after some code review Comment out something so branch compiles Added weightfiles Some code cleaning (as in ID part of cms-sw#23743) Adapt GBRForestTools exception handling to 94X Fall17 cutbased Electron ID V2 (as in cms-sw#23477) Added Photon MVA weight files Technical Egamma MVA improvements (as in cms-sw#24131) Fall17_94X_v2 MVA Photon ID Fixing mvaClassName in egammaObjectModificationsInMiniAOD_cff.py Inclusion of V2 IDs in MiniAOD fix worst charged iso variable adding photon MVA V2 to the pat::Photon Fixed copy-paste typo in Spring15 weight files New MultiTokens (as in cms-sw#24312 cms-sw#24423) cache AOD/miniAOD values to speed up repeated computations Egamma IDs back to global cache (as in cms-sw#25101) Cutbased photonID Fall17 V2 update phoID v2 cutbased Renamed cut based ID and put in MiniAOD Modified PhotonNtuplizer to have genMatch information Added variables from Fall17 ID to PhotonMVANtuplizer New MVAVariableHelper for extra MVA variables Moved Photon MVA to new interface adapted constructors to run MVA estimator in python fwlite ElectronMVA can be used in FWlite Egamma Python wrapper class for Electron MVAs Got rid of ElectronMVAVariableHelper Adapted PhotonMVANtuplizer Combined Electron Ntuplizer config files Fixing problems detected while testing exposing category in FWLite to be able to apply working point later on infrastructure to take working points in python+fwlite. implemented for noiso v2 only so far working points for the other mvas. somthing fishy with GP v1 Spring16 WPs with logistic transform
This pull request contains:
Link to slides of latest update (30th of May 2018):
https://indico.cern.ch/event/732971/contributions/3022843/attachments/1658685/2656462/eleIdTuning.pdf
Earlier updates preceding the final update:
https://indico.cern.ch/event/697079/contributions/2968124/attachments/1632990/2604194/eleIdTuning.pdf
https://indico.cern.ch/event/697576/contributions/2940576/attachments/1620927/2578913/eleIdTuning.pdf
@Sam-Harper
@lsoffi
@guitargeek