Responsible
Main.JohanSebastianBonilla

Substructure Top Tagger (Part of Boosted Jet Taggers)

Boosted Jet Taggers: Official JSS Tagger Package

Introduction

In an effort to simplify the end-user interface for the various boosted object taggers supported by the JSS subgroup, all supported taggers are built into a common environment, IJSSTagger. The new interface is based on BoostedJetTaggers-00-00-25/ and is intended to be a collection of all ATLAS-supported taggers. These taggers are designed to identify the sources of hadronic decays in the ATLAS detector (W/Z/Higgs bosons, top quarks, q/g discrimination, etc.). The methods developed and supported include, but are not limited to:

• DNN (high-level & low-level variables)
• BDT
• 'basic' MVA (2 variable taggers, typically mass+substructure)
• Shower Deconstruction
  Currently all necessary files are saved in the (BoostedJetTaggers) package, but as development continues, some of these data files (.xml, .json, .dat, .root) will be moved to the central area (SVN). Update to follow.
  A JIRA page exists for the discussion of methodologies and conventions to be applied for all ATLAS taggers.
  IJSSTagger environment is used by the JSS Tagger Package containing the following boosted object taggers: Hbb Tagger, W/Z Tagger, top Tagger, NN Tagger, and BDT Tagger. This twiki will descibe how the SubstructureTopTagger is used in the IJSSTagger environment.

The JSS Tagger Interface

To unify the structure of all jet substructure taggers, a base class (interface) has been defined: IJSSTagger, which inherits from IJetSelector.

This interface ensures that all tagging algorithms follow the same structure (no matter who writes them) and users know what to expect from any tagger they implement.
The following functions are defined by the interface:

Function	Description
StatusCode initialize()	Initialize all of the attributes of the tagger, setup of files to access and working points
int isTagged( xAOD::Jet jet )	Get the tagging result: A return value of 0 means that all cuts PASSED
StatusCode finalize()	clear or delete anything necessary

The initialize() function is called once per tagger instance, along with the constructor and its arguments. The result() function is called in the execute part of your analysis and should be called once per large-R jet. The finalize() functions is used in the finalize portion of the analysis and cleans residual memory from the IJSSTagger instance.

Checking Out the BoostedJetTaggers Package

• Current package located at
  
  • git clone https://:@gitlab.cern.ch:8443/JSSTools/BoostedJetTaggers.git

• Header include:

  #include "BoostedJetTaggers/IJSSTagger.h"

• Other package dependencies
  • Are there? Or just these two things?

  • What about the ASG thing?

Usage Recommendations 2016

Warning: Can't find topic Sandbox.BoostedJetTaggingRecommendation2016

Top Tagger Information

This tagger is recommended for use on AntiKt10LCTopoTrimmedPtFrac5SmallR20 jets, with pt > 300GeV, and |eta| < 2.0

Initializing the Tool

In addition to initializing the IJSSTagger instance using the IJSSTagger::initialize() function, the TopTagger constructor should also be called. It has three forms of the function:

• SubstructureTopTagger *t = configSubstTagger(tagName (string), taggingShortCut (string));
  • tagName is the name of the tool and also what the tool will set as an attribute to tagged jets
  • taggingShortCut can be one of the following keys
    *

tagName	Description
SmoothCut_50	pT dependent cut maintaining ~50% signal efficiency
SmoothCut_80	pT dependent cut maintaining ~80% signal efficiency
FixedCut_LowPt_50	50% efficiency in pT < 500 [GeV] region
FixedCut_LowPt_80	80% efficiency in pT < 500 [GeV] region
FixedCut_HighPt_50	50% efficiency in pT > 500 [GeV] region
FixedCut_HighPt_80	80% efficiency in pT > 500 [GeV] region

• • Ex: SubstructureTopTagger * t = configSubstTagger("LooseTopTag", "SmoothCut_80");
• SubstructureTopTagger *t = configSubstTagger(tagName (string), cutdescs vector<string>);
  • tagName is the name of the tool and also what the tool will set as an attribute to tagged jets
  • cutdescs is a vector of strings, where each string describes a cut in the form "12.34<Att<45.67" OR "12.34<Att" OR "Att<45.67". The cuts will be applied using a logical AND.
    • Att can be a jet attribute ot one of the known keys:
      • "abs(eta)"
      • "Tau32_wta" (n-subjettiness ratio)
      • "Tau21_wta" (n-subjettiness ratio)
    • The string entry can also be one of the following available shortcuts

    • tagName	Description
      SmoothMassCut_50	pT dependent cut maintaining ~50% signal efficiency
      SmoothMassCut_80	pT dependent cut maintaining ~80% signal efficiency
      SmoothTau32Cut_50	pT dependent cut maintaining ~50% signal efficiency
      SmoothTau32Cut_80	pT dependent cut maintaining ~80% signal efficiency
      SmoothMassCut_50_loweta	pT dependent cut maintaining ~50% signal efficiency. Optimized for jets in |eta| < 1.0 region
      SmoothMassCut_50_higheta	pT dependent cut maintaining ~50% signal efficiency. Optimized for jets in 1.0 < |eta| < 2.0 region
      SmoothMassCut_80_loweta	pT dependent cut maintaining ~80% signal efficiency. Optimized for jets in |eta| < 1.0 region
      SmoothMassCut_80_higheta	pT dependent cut maintaining ~80% signal efficiency. Optimized for jets in 1.0 < |eta| < 2.0 region
      SmoothTau32Cut_50_loweta	pT dependent cut maintaining ~50% signal efficiency. Optimized for jets in |eta| < 1.0 region
      SmoothTau32Cut_50_higheta	pT dependent cut maintaining ~50% signal efficiency. Optimized for jets in 1.0 < |eta| < 2.0 region
      SmoothTau32Cut_80_loweta	pT dependent cut maintaining ~80% signal efficiency. Optimized for jets in |eta| < 1.0 region
      SmoothTau32Cut_80_higheta	pT dependent cut maintaining ~80% signal efficiency. Optimized for jets in 1.0 < |eta| < 2.0 region
  • Ex 1: SubstructureTopTagger * t = configSubstTagger("MassAndTau32Tag", {"120000<m", "Tau32_wta<0.61"} );
  • Ex 2: SubstructureTopTagger * t = configSubstTagger("SmoothMTag", {"SmoothMassCut_50", "SmoothTau32Cut_50"} );
• SubstructureTopTagger *t = configSubstTagger(tagName (string), cutdescs vector< vector<string> >);
  • tagName is the name of the tool and also what the tool will set as an attribute to tagged jets
  • cutdescs is a vector of vectors of strings, where each vector of strings, the string entries are treated as described in the previous bullet point and the cuts are applied to each vector of strings using a logical AND. The separate vectors of strings will then be combined using a logical OR.
  • Ex:SubstructureTopTagger * t = configSubstTagger("MassAndTau32Tag",
    { {"abs(eta)<1","120000<m", "Tau32_wta<0.61"} ,
    {"1<abs(eta)","120000<m", "Tau32_wta<0.61"} } );

Tool Use, Jet Decoration, and Return Value

In the execute() portion of your analysis, the TopTagger tool that was initialized can be used. Inheriting the same environment as the other taggers in the BoostedTJetTaggers package, the tagging results can be obtained by calling the isTagged( xAOD::Jet jet ) function.

Example:

returnValue = m_SubstructureTopTagger.isTagged( xAOD::Jet jet ); // run for each large-R jet

returnValue is an integer (0 or 1) denoting whether the large-R jet used as an argument is tagged or not. The isTagged() function may also decorate the argument jet. Is this true for for SubstructureTopTagger?

Samples

Also took this from the old twiki. Are these up to date?

First look at available MC15 samples by Lily here

Status for the Top-Tagging samples here

For the signal: ttbar_allhad or ttbar_nonallhad in mtt slice, for several variations (in DC14 and MC15 for some variations)
For background: JZ*W samples (similar to W-tagging).

Signal samples

Tasks within SubstructureTopTagger

(I took this straight from the old twiki)

Editors
contributors : Julien Caudron, James Ferrando

Samples preparation
progress : started
contributors : Julien Caudron (+ Chris Delitzsch, Lily Asquith)
comments :

• Make sure that all the needed samples are available or requested.
  Simple Top Tagger variables optimisation, ROC curve, WP definitions
  progress : preliminary results
  contributors : Dortmund, Grenoble
  comments :
• Dortmund contribution using 14TeV NTUP_COMMON
• Grenoble study in xOAD
• Pile-up optimisation : also done by Dormund. In this case, it's done with specific high pile-up samples, we can probably add this without update to xAOD
• Do we want to revisit the choice of variables according to their uncertainties ? (i.e., to choose a variable that is less efficient to reject background but that is less sensible to the modeling)
  Simple Top Tagger variables uncertainties
  progress : preliminary results
  contributors : Grenoble, Chris Delitzsch, James Ferrando
  comments :
• Contributions from Grenoble, to be done with the correct samples, for the correct variations (work in progress)
• Run-1 uncertainties for relevant variables (Chris and James)
  Simple Top Tagger scale factor and scale factor uncertainty
  progress :
  contributors :
  comments :
  Negligible variations cross-check
  progress :
  contributors : need contribution
  comments :
• Low priority ?
• Few variations will not be included in the whole process because we expect their effect to be negligible. Only few samples exist, and we can check that, indeed, those variations are negligible.
  Mass calibration for large-R jet
  progress :
  contributors : shared with W-tagging ?
  comments :
• currently, the text will be written in the top-tagging paper repository, but it can move to the w-tagging paper if it's more relevant
  Test of the Simple Top Tagger with the ttH signal
  comments :
• proposed by the boosted ttH group, will be studied here.
  -- JohanSebastianBonilla - 2016-11-24