Skip to content

Latest commit

 

History

History
337 lines (261 loc) · 8.06 KB

AnalysisPlan.md

File metadata and controls

337 lines (261 loc) · 8.06 KB
Raw

Analysis Plan

TODOs

Incorporate remaining systematics

Skimmer

Shapes / Values

  • Pileup
  • ParticleNet Xbb
    • Split up uncertainties
  • JES/R
  • JMS/R
  • Trigger SFs
    • Stat. Unc.
    • Correlated Syst.
  • Top pt
  • Theory
    • pdf uncertainties
    • parton shower weights
    • scale variation
    • W k factor

Datacard

  • MC Stats
  • Lumi
  • Pileup
  • Trigger SFs
    • Stat.
    • Correlated Syst.
  • ParticleNet Xbb
    • Separate uncertainties
  • JES/R
  • JMS/R
  • Top pT
  • Theory
    • BR
    • pdf uncertainties
      • rate
      • shape?
    • QCD scale
    • alpha_s (for single Higgs)
    • parton shower weights
    • scale variation?
    • W k factor??

Update processor

  • New e, mu, b-tag jets selections from VHbb
  • Re-run with VV regressed mass for Dijet variables
  • FatJet ID
  • Look at ID scale factors
  • New LP Method

New LP SFs

  • Update LP method

Nonresonant

  • BDT ROC
    • Try multi-class BDT
    • Try equalizing background weights
    • Optimize hyperparams
    • Trim features
    • Check sculpting
  • Scan Txbb, BDT WPs
  • Run over all kL and k_2V
    • Re-run VBF with gen selection
  • HH inference
  • Theory uncertainties

Statistical tests of fits

  • GoF
  • F-test
  • Impacts

Resonant X->HY

  • Control plots
  • Preliminary signal region
  • Validation region
  • 2D fit, xUL
    • GoF
    • F-test
  • Run over all signals
    • TRSM, NMSSM exclusions
  • Scan Txbb, THWW working points

Semi-leptonic ttbar selection skimmer

Resolved WW veto

  • Test veto on 2 AK8 W-tagged jets

TTbar corrections

May not be necessary given low yield

  • Tagger efficiency
  • Recoil
  • JMS
  • Regressed mass
  • BDT
  • Check VBF?

Plan

Feb 13 - 17

Raghav:

  • Trigger efficiencies
  • Add to skimmer:
    • num e, mu
    • e, mu 4 vectors
    • ak4 b jets (medium btag) - outside bb jet
    • check control plots on num_x
  • Check vetoes on these

Cristina

  • Systematics

Feb 20 - 24

Raghav:

  • 2017 fits
    • PR: Systematics into datacard

Cristina:

  • update xsecs
  • Goodness-of-fit, f-test, impacts

Feb 27 - Mar 3

Raghav:

  • Full run 2, all samples

Mar

  • Wrap up nonresonant for now with full run 2 limits and GoF
  • Develop resonant analysis strategy
  • Plots, selections for subset of signals
  • Get 2D fit working

Apr

  • Get 2D fit working
  • Upper limits for subset of signals
  • Upper limits for all signals!
  • Complete v1 of AN
  • Higher mass ParT training samples

Apr 24 - 28

  • Complete v1 of AN
  • B2G Workshop talk
  • F-test
  • Start WP Scan

Jun 5 - 9

  • Update semi-leptonic ttbar processor
  • Twiki
  • Respond to comments
  • Samples
    • Nanogen
  • Skimming VBF
  • Running over kL, k2V

In progress:

~Completed:

Preliminary 2017 cut-based signal and background yields estimate

  • using a coarse-grained grid search on pT, msd, and tagger scores
  • 2017 lumi + data only
  • measured for two AK8 fat jets, two AK15 fat jets, and a hybrid (AK8 for bb candidate, AK15 for VV candidate)
  • background estimation from data sidebands
  • with AK8 and AK15 mass-decorrelated ParticleNet Hbb taggers + NOT mass-decorrelated AK8 ParticleNet H4q tagger

Trigger scale factor measurements

  • Measured for AK8, AK15, and hybrid jets, single-jet 2D (mass, pT binned) efficiencies (applied assuming prob. of each fat jet passing trigger is independent) and 3D (jet 1 mass, jet 1 pt, jet 2 pt binned) efficiencies (processors)

  • Decided on AK8 only - ~same sensitivity, significantly easier practically

  • Measure for all years

  • Update selection

  • Check if binning in VV tagger is necessary (probably not since only btag is in the trigger)

  • Investigate high unc.

Processor for skimming nano files

https://github.com/rkansal47/HHbbVV/blob/main/processors/bbVVSkimmer.py

  • Currently includes:
    • Signal gen-matching cuts
    • Pre-selection kinematic cuts
    • Inference via triton server running on SDSC
    • Lund plane scale factors for skimmer
    • Saving flat skimmed data to parquet files, and metadata (total events, cutflow) to pickles

Update 2/23

  • FatJet selections
  • Modify JECs code to save only variations of pT
  • Add JMS/R
  • Regressed mass cut
  • Add e, mu, b-tag jets
  • Add tagger vars
  • Dijet variables

Triton/SONIC inference server

https://github.com/rkansal47/sonic-models https://gitlab.nrp-nautilus.io/raghsthebest/triton-server

Server for running inference with our new HWW tagger on samples.

BDT Training

https://github.com/rkansal47/HHbbVV/blob/main/src/HHbbVV/scripts/pickle_scripts/TrainBDT.py

Tagger

HWW tagger + mass regression development (Zichun's tagger repo and Ish's regression repo)

Fits, combine

Full Run 2 and all UL samples

Lund plane scale factors

  • Implemented and validated for top jets in control region
  • Implemented and measured for nonresonant signal

Post-processing

Update post-processing

  • Check control plots for all years
  • Re-train BDT for all years
  • Templates, systematics for all years
  • Update datacard with all years