Implementation of top reconstruction methods

[ywkao]

Hadronic channel

Two kinds of chi-2 methods are implemented for handling the two kinds of FCNH signals (ttbar signal and single top signal). They are based on 3x3 covariance matrix and 2x2 covariance matrix respectively (these matrices are calculated from 2017 signal samples with MC truth-matching information). The methods aim at picking up the jets that are more likely coming from t->qH, and t->bW(jj) processes. In practice, these jets (b, j, j, q) are determined by the combination with the least chi-2 value and are manipulated in terms of the indices of the list of selected jets. Particularly, the b-jet in the process t->bW is simply determined by the jet with the highest b-tag scores because currently it is found to have the best matching efficiency (~70%) according to MC truth information.

Leptonic channel

The method with a quadratic equation is implemented for the calculation of neutrino Pz. This method assumes a fixed value of W boson mass (80.379 GeV) and solves an equation based on the kinematics of W->lv. In addition, the top quark (t->bW) is reconstructed subsequently with the b-jet determined by the jet with the highest b-tag scores (matching efficiency of the b-jet coming from hard process ~ 80%). Besides, for ttbar signal, the jet from t->qH coupling is determined by a simple chi-2 sorting method; in other words, the combination of jγγ with the least chi-2 value (closest to top quark mass, 173 GeV) is chosen (matching efficiency of the jet ~ 40%).

Enumeration of some intriguing variables

The kinematics related to both top quarks and Higgs boson may offer some features that background events don't have. These variables include the open angles and the ptOverM of the top quark that decays through an FCNH process.

For both ttbar and single top signal

• chi2_tbw_deltaRdipho

For ttbar signal only

• chi2_3x3_tbw_deltaRdipho
• chi2_3x3_qjet_deltaRdipho
• chi2_3x3_tqhptOverM
• chi2_3x3_tqh_deltaRtbw

• chi2_3x3_tqh_deltaRdipho

  More variables can be found in the files:

  Loopers/MakeMVABabies_ttHLeptonic.h Loopers/MakeMVABabies_ttHHadronic.h

[sam-may]

Thanks for making the PR, @ywkao . I will merge it today.

Some notes:

• I plan to deprecate the various prep.py, prep_dnn.py, etc scripts and have a single prep script that is easily configurable for prepping hdf5 files for BDT/DNN training with a specified list of input features. This should make things like comparing top tagger performance easier to do from the command line.