wmass / cmgtools-lite Goto Github PK

Since we are using free functions multiple times, especially the electron smearer which has a slo TRandom inside, implement as much as possible in friend trees. A possible list:

1. electron corrected pt
2. trigger SF
3. pass full ID+iso+conv rejection bit

Features to be added from discussion with @cippy:

1. Cross Lepton-Jet cleaning (favouring the lepton)
2. PU weights for different sections of 2016 data-taking (eg RunB-F)
3. kinematic variables for Z pT reweighting (lower priority)

In case we keep using the json on electron dataset (30.9/fb instead of 35.9) to force L1 pt thresholds below or equal to the HLT ones, we should probably remake the PU reweighting

I would expect that data selected by the json has lower PU than the total dataset.

• ele scale: should be using the data/MC discrepancy of the residual corrections
• fake shape systematics had an obsolete reweighting (not the super duper latest smooth one)

Skims based on already produced ntuples, with a reasonable compression, takes ~1day with full 2016 MC and SingleElectron dataset of 2016. Need to parallelize.

To be added:

1. symmetrization of the NNPDF systematics
2. update of mergeCardComponentsAbsY to account for the PDF systematics

(@mdunser just to remind me what is still needed to do...)

• check the SingleElectron triggers in 2016 BF
• check if the SingleElectron can be used in G-H

@crovelli can you look at it?In case we cannot use G-H the lepton SFs have to be reverted to ICHEP 2016 ones from this twiki

• Check bin-by-bin migrations in T&P
  • compare eff binned in GEN-PT vs eff binned in TRUE-PT
• make a 1D plot of systematics
  • 1 plot of syst vs YW for 2 polarizations
  • 1 plot of syst vs eta-pt unrolled
• Efficiency:
  • add a systematic on the signal modelling due to FSR, QCD modelling on the Z lineshape, etc (see differential DY)
• Wtaunu
  • correlate PDFs with signal and treat the WTau as signal?
  • or reweight the WTau by a moderate variation of pdf13 (or one of the mostly constrained PDF parameters)
• MET modelling:
  • smear the met template in one eta/pt bin and repeat the fakerate estimate

PDFs, qcd scales, etc.

Reporting Josh's suggestion:

What I described above is actually more or less equivalent to
"QED-jets", since additional leptons may be included together with
photons as part of the dressing.

This may be more algorithmically convenient to implement as:

1. Sort all isPromptFinalState photons and charged leptons by pt.

2. Starting from the highest pt particle, form QED-jets in a cone
   around the seeding particle, removing all included particles from the
   list of seeds.

3. Continue until there are no more particles left to seed QED-jets.

Something like this is also the least sensitive to the details of the
QED FSR, including pair production.

For the neutrino one can just take the highest pt isPromptFinalState
neutrino as said.

do cleanly in CMSSW, but can also be tried on ntuples (in case we need to recompute it)

@emanueledimarco
@crovelli

Hi,
I just noticed that in the new friend trees here [0] we have the puWeight variable for PU reweighting.
However, it is also in the big ntuples.
This could be a problem when reading the tree. Which variable is going to be actually read?

I will produce a plot of Nvertices after the reweighting, but this needs to be fixed. The simplest solution is to rename the variable in the friend trees.
For the moment I only have this [1] with the Z. The normalization is totally of because I changed manually the cross section to the one reported here [2] but I didn't multiply by 3.
Yet it is clear that after the reweighting the peak is different between data and MC.

Marco

[0] /eos/cms/store/group/dpg_ecal/comm_ecal/localreco/TREES_1LEP_80X_V3/
[1] http://mciprian.web.cern.ch/mciprian/wmass/13TeV/zeePlots/TREES_1LEP_80X_V3/HLT_SingleEl/withSF/EBEB_Zamc@NLO/nVert.pdf
[2] https://twiki.cern.ch/twiki/bin/viewauth/CMS/StandardModelCrossSectionsat13TeV

1) check tau polarization with LO sample (need new trees)

2) reweight Z-pT to SMP-17-010

3) implement stat smoothing

4) inflate eff uncertainties

-> do this once the nominal templates are orthogonal

5) start at pT 30 for muons. potentially coarser pT binning

-> submit jobs before starting the weekend

6) look at what happens in single charge fits

-> did this test a while back. didn't change a thing.

7) look at A4 as well (with fully floating we can already)

-> fully floating we have already. add to slides?
-> for long fixed the datacards are now with josh

8) redo the expected cross sections with orthogonal samples

9) de-correlate the QCD scales between the helicities

-> all three!

A reminder that the pt-dependent FR shape systematics have been removed and should be put back following the new parameterization. See #269 (comment)

Given the non trivial efficiency of single electron triggers as in Afiq's presentation as a function of eta:

we need to apply scale factors for L1 x HLT efficiency on top of the offline selection scale factors. This is also time dependent, since the menu and the isolation corrections changed.

• rebinning of the templates? necessary?
• fix the elescale naming scheme somehow (still not sure what's wrong)
• qcd scale for Z, add back non-numbered and make uncorrelated with W
• charge asymmetry
• make the pT slope FR uncertainty uncorrelated in eta (maybe?)

To avoid steps in the final distribution