Tau POG short exercise @ CMSDAS 2019

The esercise consists of two main parts: first produce flat ntuples containing quantities relevant for tau studies, then process these ntuples to produce performance plots.

You can find the indroductory slides at this link

Outline

• installation
• package content
• how to run the ntupliser on DY and QCD MC
• investigate the ntuples, produce performance plots
• links and references

Installation

The present package does not depend on CMSSW, strictly speaking: basically we only need a working version of ROOT and pyROOT, plus a few further python packages. However, ROOTis not available as standalone on gridui{1,2} therefore it is convenient to just rely on CMSSW, which comes with everything we need. In any case, we choose a very recent release CMSSW_10_2_10.

cmsrel CMSSW_10_2_10
cd CMSSW_10_2_10/src
cmsenv
git clone https://github.com/rmanzoni/TauCMSDAS19.git
cd TauCMSDAS19

If you'd like to contribute to this package (and you're more than welcome to do so!) you may want to first fork this repository, then replace

git clone https://github.com/rmanzoni/TauCMSDAS19.git

with

git clone https://github.com/YOUR_GITHUB_USERNAME/TauCMSDAS19.git

in the instructions above. You can submit your changes to this package as pull requests.

Package content

In the TauCMSDAS19 directory you'll find the following files:

• read_taus_nano.py is the ntupliser, you'll have to run this python script to produce your own ntuples (details in the next section)
• treeVariables.py where the branches of the output flat ntuples are defined. Here you can adjust the event content to your taste, e.g. adding more information into the flat ntuple, as well as you can see how the information is fetched from the original NanoAOD
• files.py and deltar.py are utils to fetch the original NanoAODs and to compute deltaR repectively. Ideally you won't need to touch them at all
• branches.txt is a snapshot of the documentation of the NanoAOD content. Very useful if you want to know what's what
• tau_plotting.ipynb jupyter notebook with step by step comments and code snippets to produce performance plots

Run the ntupliser

The ntupliser reads NanoAOD files stored on the shared file sistem in Pisa at /gpfs/ddn/cms/user/cmsdas/2019/TauExercise/{dy, qcd} and produces flat ntuples where each reco tau makes an entry.

Side note: these samples are a (partial) local copy of these, normally published and hence accessible from anywhere, QCD and DY samples.

The choice to read off NanoAOD is based on different reasons:

• it is a lightweght, centrally produced (and validated) format
• Tau POG is moving towards larger adoption of NanoAOD, especially for performance measurements

Everything is already setup for you and, in principle, you just need to push a button, however you're encouraged to dig into the code (it's very straightforward and hopefully well documented) and try to understand the basics.

In short, this is all you need to produce the ntuple

ipython -i -- read_taus_nano.py --dy --maxevents 1000000 --logfreq 1000
ipython -i -- read_taus_nano.py --qcd --maxevents 1000000 --logfreq 1000

Notice that eventually you'll need to produce two ntuples, one from DY, to have a sample of genuine taus, and one from QCD, to have a large sample of jets faking taus.

Performance plots

A jupyter notebook are guided step by step towards producing a few performance plots which are typical Tau POG bread 'n' butter: efficiencies, fake rates, ROC curves and more.

You can visualise the content of the jupiter notebook at this link https://github.com/rmanzoni/TauCMSDAS19/blob/master/tau_plotting.ipynb

The notebook contains comments, questions, instructions and the code to produce the plots. During the exercise we'll gothrough the notebook together and you'll be encouraged to find your own code implementation to satisfy the requests.
You can always look at the official code snippets, they're in the notebook itself, but only do it as last resort and give yourself a chance to learn by trying in first person!

We suggest that, during the hands on exercise, you follow the passages in the notebook interactively in a ipython session and/or write code in a separate python file.

Tips & Tricks

• Let the job run even if the session is closed:
  • https://en.wikipedia.org/wiki/Nohup
  • https://en.wikipedia.org/wiki/GNU_Screen
• reco decay mode logic: decayMode = 5*(n_ch-1) + n_pi0
  • e.g.: 1-prong + 1pi0 ==> 5*(1-1) + 1 = 1
• python debugger: https://docs.python.org/2/library/pdb.html
  • insert this line import pdb; pdb.set_trace() at the exact point you want to pause the script execution and interactively inspect the code

Links and references

Tau POG on indico
https://indico.cern.ch/category/6330/

Main Tau POG TWiki
https://twiki.cern.ch/twiki/bin/view/CMS/Tau

Recipes and recommendations
https://twiki.cern.ch/twiki/bin/viewauth/CMS/TauIDRecommendation13TeV

Software guide
https://twiki.cern.ch/twiki/bin/view/CMSPublic/SWGuidePFTauID

Tau long exercise at 2017 CMS Physics Object School
For a more extensive plunge in Tau realm, including accessing to the information stored in MiniAOD, rerunning the HPS algorithm and more, see:
https://twiki.cern.ch/twiki/bin/view/CMS/SWGuideCMSPhysicsObjectSchoolBARI2017Tau
https://github.com/rmanzoni/TauRecoCMSPOS