Electron and Photon Regressions in RECO (76X)

[lgray]

This PR implemented the latest regressions in RECO, they are no longer run in MiniAOD.

The Photon and Electron dataformats are extended to now save the extra set of commonly used shower shapes for ID and regression, saving space at the MiniAOD level since we are dropping strings.
full5x5 and fractionize shower shapes are saved.

Isolations are still embedded in MiniAOD. (it is WIP to put them directly in RECO)

Expected changes in electron and photon energies in AOD.
Validation plots posted below.

@ikrav VID functions in all respects in this PR now.

@bendavid @gpetruc

[lgray]

@cmsbuild please test

[cmsbuild]

The tests are being triggered in jenkins.

[cmsbuild]

A new Pull Request was created by @lgray (Lindsey Gray) for CMSSW_7_6_X.

Electron and Photon Regressions in RECO (76X)

It involves the following packages:

CommonTools/CandAlgos
DataFormats/EgammaCandidates
PhysicsTools/PatAlgos
RecoEgamma/EgammaElectronAlgos
RecoEgamma/EgammaElectronProducers
RecoEgamma/EgammaPhotonAlgos
RecoEgamma/EgammaPhotonProducers
RecoEgamma/EgammaTools
RecoEgamma/ElectronIdentification
RecoEgamma/PhotonIdentification
RecoHI/Configuration
RecoParticleFlow/Configuration

@cmsbuild, @cvuosalo, @vadler, @monttj, @slava77 can you please review it and eventually sign? Thanks.
@TaiSakuma, @rappoccio, @echapon, @mmarionncern, @imarches, @makortel, @acaudron, @dgulhan, @jdolen, @ferencek, @mandrenguyen, @yetkinyilmaz, @Sam-Harper, @nhanvtran, @schoef, @richard-cms, @MiheeJo, @jazzitup, @ahinzmann, @yenjie, @kurtejung, @gpetruc, @istaslis, @mariadalfonso, @pvmulder, @bachtis this is something you requested to watch as well.
You can sign-off by replying to this message having '+1' in the first line of your reply.
You can reject by replying to this message having '-1' in the first line of your reply.
If you are a L2 or a release manager you can ask for tests by saying 'please test' in the first line of a comment.
@Degano you are the release manager for this.
You can merge this pull request by typing 'merge' in the first line of your comment.

[lgray]

@richard-cms Does HI run the PF relinking after particleFlowTmp? I ask since the regressions are now applied in the finalized photon and electron producers that are run after PF, and did not see them being customized for HI.

[harmonicoscillator]

Matt Nguyen is currently investigating running PF re-linking for HI. Our
initial tests showed that everything is working fine, but re-linking is not
actually enabled right now.

Alex

On Fri, Sep 18, 2015 at 1:38 PM, Lindsey Gray [email protected]
wrote:

@richard-cms https://github.com/richard-cms Does HI run the PF
relinking after particleFlowTmp? I ask since the regressions are now
applied in the finalized photon and electron producers that are run after
PF, and did not see them being customized for HI.

—
Reply to this email directly or view it on GitHub
#11367 (comment).

[harmonicoscillator]

@mandrenguyen

[cmsbuild]

+1
Tested at: a8bd036
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-11367/8213/summary.html

[cmsbuild]

Comparison is ready
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-11367/8213/summary.html

[lgray]

@cmsbuild please test

[cmsbuild]

The tests are being triggered in jenkins.

[cmsbuild]

Pull request #11367 was updated. @cmsbuild, @cvuosalo, @vadler, @monttj, @slava77 can you please check and sign again.

[cmsbuild]

Pull request #11367 was updated. @cmsbuild, @cvuosalo, @vadler, @monttj, @slava77 can you please check and sign again.

[lgray]

@cmsbuild please test

[cmsbuild]

The tests are being triggered in jenkins.

[cmsbuild]

+1
Tested at: 63708e9
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-11367/8442/summary.html

[cmsbuild]

Comparison is ready
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-11367/8442/summary.html

[lgray]

Did one final check: for 2.25 TeV Z-primes the regressed resolution is better by 8%, the regressed scale is worse by 1%. This is acceptable for general use.

[slava77]

there is now an improved ("pure") egamma performance as pfCandidates:

• ele35 has higher peak and more centered mean (new and ref are flipped here: black is "new")
  
• single gamma10 electrons pfCandidates are also closer to 10 GeV (new and ref are flipped here: black is "new"):
  
• single gamma10 photons pfCandidates look slightly worse 10 GeV, I suppose this is just edge effects (new and ref are flipped here: black is "new"):
  
• gamma 35 PF candidate electrons are a bit better, look more symmetric around 35 (new and ref are flipped here: black is "new"):
  
• gamma 35 PF candidate photons are a better on the resolution side (new and ref are flipped here: black is "new"):
  
• ele 1000 are also slightly better as pfCandidates [manual fit gives mean pt/gen-1 move from -0.16% to -0.13%; sigma from 1.123% to 1.107%](new and ref are flipped here: black is):
  

Comparing to the previous iteration: the regression1Energy* in photons is now set to the values from regression2Energy*, which triggers changes in the plots.

Comparing with the original and new pfcandidates and egamma candidates: the consistency in momentum assignment is now restored.

[lgray]

@Slava7 10 GeV photon "worse" resolution is a convolution of edge effects + using all pf candidates (rather than those that are selected as real photons). This makes sense to me.

[slava77]

+1

for #11367 63708e9

• changes in the code are visually OK, implement what's declared in the PR:
• tested locally in CMSSW_7_6_X_2015-09-22-1500
  • egamma candidates energies are updated (resolution improves), corresponding regression uncertainties change too (see plots from the first iteration)
  • pfCandidates corresponding to the egamma candidates change their energies consistent with the egamma candidate
  • new shower shape variables show up in the outputs as expected (no particular quality control of this was done from my side other than spot-checking the values are non-zero with some sensible distributions)
  • event size increases negligible at RECO and AOD level (under 0.5%); miniAOD size on 200 events goes up by 2% (this should decrease with more input events due to compression).
  • monitored physics variables change if related to the egamma candidate energies (reco::GsfElectron and reco::Photon GED collections) or come downstream of PF candidates
    • jets are not affected significantly for regular q/g jets
    • egamma candidates (reco::GsfElectron and reco::Photon GED collections) now have a better response (energy over gen) as expected
• jenkins tests pass and comparisons with baseline are consistent with the more detailed tests

[slava77]

@monttj please check this PR, the AT signature is needed.
Thank you.

[monttj]

+1

[cmsbuild]

This pull request is fully signed and it will be integrated in one of the next CMSSW_7_6_X IBs (tests are also fine). This pull request requires discussion in the ORP meeting before it's merged. @davidlange6, @Degano, @smuzaffar

[davidlange6]

+1