GlasgowBoostedTTH

Limits (at 35fb^-1)

	boosted only stat	boosted only syst	resolved-minus-boosted stat	resolved-minus-boosted syst	combined stat	combined syst
3311 vanilla	3.01209	-	0.673451	-	0.665444	-
3211 vanilla	10.80678	-	0.717076	-	0.695144	-
3311 1b,ge1nb	1.97755	-	0.678389	-	0.622131	-
3211 1b,ge1nb	3.38834	-	0.689422	-	0.64098	-
3211 1b,ge1nb (sarah xcheck)	3.38834	-	0.689422	-	0.64098	-
3311 1bgt1nb 300GeV	N/A	N/A	N/A	N/A	0.616	N/A
RESOLVED	N/A	N/A	N/A	N/A	0.628686
RESOLVED (sarah xcheck)	N/A	N/A	N/A	N/A	0.628687

ICHEP stuff

Running TMVA

• TMVA Core area : /afs/cern.ch/user/s/scrawley/public/TMVA_Core
• TMVA Inputs : /afs/cern.ch/user/s/scrawley/public/TMVA_Inputs
• TMVA Training Outputs : /afs/cern.ch/user/s/scrawley/public/TMVA_Outputs

To run the training, navigate to the core area and copy a TMVAClassification.C file from the Outputs area if one is not already present in the training area.

"root -l TMVAClassification.C" to run the training.

Changes can be made to the training by editing the TMVAClassification.C files.

• The output root file can be specified at the line : TString outfileName( "EXAMPLE.root" );
• The variables used in the training are specified with the line : factory->AddVariable( "VARIABLE_NAME", 'DATA TYPE (F(float) or I(integer)' );
• Signal root file specified with : TFile *inputS = TFile::Open("EXAMPLESIGNAL.root");
• Background root file specified with : TFile *inputB = TFile::Open("EXAMPLEBKG.root");
• Signal and Background weights can be specified with : factory->SetSignalWeightExpression ("FinalWeight"); or factory->SetBackgroundWeightExpression ("FinalWeight");
• The BDT Settings can be set here :

if (Use["BDT"]) // Adaptive Boost

factory->BookMethod( TMVA::Types::kBDT, "BDT",

"!H:!V:NTrees=400:MinNodeSize=2%:MaxDepth=3:BoostType=AdaBoost:AdaBoostBeta=0.5:UseBaggedBoost:BaggedSampleFraction=0.5:SeparationType=GiniIndex:nCuts=20:IgnoreNegWeightsInTraining=True" );

After running the BDT, the separation should be shown in the terminal printout, and a GUI will pop up with options to draw some plots. So far plots that we have been using are produced by clicking:

• (1a) Input Variables (training sample) - Separation plots for the variables used in training.
• (3) Input Variable Linear Correlation Coefficients - Correlation Plots for Signal and Background
• (4b) Classifier Output Distributions (test and training samples superimposed) - Overtraining plot with KS test.
• (5b) Classifier Background Rejection vs Signal Efficiency (RoC Curve)

If the overtraining plot fails and crashes, the weights provided to the BDT in TMVAClassification.C need to be changed so that negative weights are set to 0, or positive.

Plots drawn will be stored in the plots/ directory in TMVA_Core. Weights will be in the weights/ directory.

To-do lists

• Modelling systematics
• Fits for 6 and/or 10fb-1

Current plots

With systematics:

* 3211 https://sboutle.web.cern.ch/sboutle/ControlPlots_boosted_emujets_3incl_3incl_1incl_1incl/

Without systematics:

* 3311 https://sboutle.web.cern.ch/sboutle/ControlPlots_boosted_emujets_3incl_3incl_1incl_1incl/

* 32ex11 http://www.ppe.gla.ac.uk/~sboutle/ControlPlots_boosted_emujets_3incl_2excl_1incl_1incl/

---

Useful Links

Derivations: https://twiki.cern.ch/twiki/bin/view/AtlasProtected/TopDerivationMC15List

PlotFactoryBoosted: https://svnweb.cern.ch/cern/wsvn/atlas-aknue/GlaProjects/PlotFactoryBoosted/

AnalysisTop: https://twiki.cern.ch/twiki/bin/viewauth/AtlasProtected/TopxAODStartGuide

This is a general thing about xAODs: https://twiki.cern.ch/twiki/bin/view/AtlasComputing/SoftwareTutorialxAODAnalysisInROOT

ttbar resonances: https://twiki.cern.ch/twiki/bin/view/AtlasProtected/TtbarResonances201