/azh_coffea

Primary LanguagePython

azh_coffea

Useful Links:

Useful Commands:

Several plotting Jupyter notebooks are available. To open a Jupyter notebook on cmslpc, you must first ssh with a -L option specifying the local host:

ssh -L localhost:8888:localhost:8888 <USERNAME>@cmslpc-sl7.fnal.gov

To start the notebook, run jupyter notebook, specifying the same port you used in your ssh command:

jupyter notebook --no-browser --port=8888 --ip 127.0.0.1

Quickstart

Most of the repo's useful contents are organized in the azh_coffea/src directory. Here's a rundown of this directory's contents:

  • azh_analysis contains the analysis code itself lives, including Coffea processors, event selection functions, and relevant utilities. It is not yet installable as a package, though it may one day be.
  • corrections contains the scale factors, fake rates, and efficiency measurements that are plugged into the analysis.
  • condor contains all the necessary scripts to submit analysis jobs to the LPC Condor cluster.
  • notebooks contains Jupyter notebooks designed to collate intermediate files, test the analysis processors, and produce plots.
  • samples contains sample lists and scripts to produce their absolte paths.

Additionally, several scripts designed to run coffea processors are available, the main one being run_analysis.py. You can test this script by running:

python run_analysis.py -s signal_UL -y 2018 --mass 225 --test-mode

The source, or -s flag, is defined to be <data type>_<legacy status>, e.g. signal_UL for ultra-legacy signal code. The year, or -y flag, denotes the data-taking era, either 2018, 2017, 2016postVFP, or 2016preVFP. The remaining flags indicate that we're running over 1 AZh signal sample with an A mass of 225 GeV.

Running with condor

We can scale up and run over the full 225 GeV AZh signal sample by navigating to the condor directory. There, we'd run the following command:

python submit.py -y 2018 -s signal_UL --mass 225 --submit

Running with lpcjobqueue

The Coffea team provides a tool called lpcjobqueue (see their repo), which is a Dask-based job queueing plugin for the LPC Condor. To run this code, we need to modify several job submission parameters; therefore, we provide a forked copy of lpcjobqueue specific to this code:

https://github.com/GageDeZoort/lpcjobqueue

To set up job submission via lpcjobqueue, you'll need to grab a copy of bootstrap.sh and execute it, creating a shell script. The shell script transfers you to the /srv directory and activates the Coffea Singularity shell.

bash bootstrap.sh
./shell
python run_distributed_analysis.py -y 2018 -s MC_UL --test-mode

Selections

The analysis selections are stored in selections/preselections.py and selections/fake_rate_selections.py. Additional utilities are available in the utils/*.py.

Triggers

Single light lepton triggers are used to identify Z-->ll decays. Trigger selections and filters are applied by functions in selections/preselections.py. The following triggers and filters are used in this analysis:

Type Year Path Filter
Single Electron 2017/18 Ele35_WPTight_Gsf HLTEle35WPTightGsfSequence
2016 HLT_Ele25_eta2p1_WPTight_Gsf hltEle25erWPTightGsfTrackIsoFilter
Single Muon 2017/18 IsoMu27 hltL3crIsoL1sMu * Filtered0p07
2016 HLT_IsoMu24 hltL3crIsoL1sMu * L3trkIsoFiltered0p09
2016 HLT_IsoTkMu24 hltL3fL1sMu * L3trkIsoFiltered0p09

The listed trigger filters are the final filters in the respective HLT trigger path. All paths and their respective filters are listed in the TriggerPaths Git Repo; given a specific year, you can search for a relevant trigger path and find all of its relevant filters.

MET Filters

MET filters are applied to rejecct spurious sources of MET, e.g. cosmic ray contamination. MET filters are applied according to the recommendations in the MissingETOptionalFiltersRun2 Twiki.

Primary Vertex Filters

The main primary vertex in each event is required to have > 4 degrees of freedom and to satisfy |z| < 24cm and \sqrt{x^2 + y^2} < 2cm.

b-Jet Filters

b-jets are required to be baseline jets passing the medium DeepFlavorJet discrimination working points listed in the BtagRecommendation Twiki. Relevant b-tag scale factor calculations are detailed in the BTagSFMethods Twiki.

Data

Samples

Samples are organized in to sample .csv files containing sample names, event counts, etc. and sample .yaml files containing sample: [file1, file2, ...] dictionaries. Sample .csv files are listed in sample_lists/ and sample .yaml files are listed in sample_lists/sample_yamls/.

  • 2018: sample_lists/data_UL_2018.csv, sample_lists/sample_yamls/data_UL_2018.yaml
  • 2017: sample_lists/data_UL_2017.csv, sample_lists/sample_yamls/data_UL_2017.yaml
  • 2016postVFP: sample_lists/data_UL_2016postVFP.csv, sample_lists/sample_yamls/data_UL_2016postVFP.yaml
  • 2016preVFP: sample_lists/data_UL_2016preVFP.csv, sample_lists/sample_yamls/data_UL_2016preVFP.yaml

Pileup Weights

Following the recommendations in https://twiki.cern.ch/twiki/bin/viewauth/CMS/PileupJSONFileforData,

  • 2018: /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/certification/Collisions18/13TeV/PileUp/UltraLegacy/
  • 2017: /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/certification/Collisions17/13TeV/PileUp/UltraLegacy/
  • 2016: /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/certification/Collisions16/13TeV/PileUp/UltraLegacy/

Pileup weights are derived from the ratio of the data pileup distribution (from the corresponding file above) to the relevant MC pileup distribution. These weights are pre-derived and queried at run-time during the analysis.

Golden JSONs

Recommended luminosity, golden JSON file information: https://twiki.cern.ch/twiki/bin/view/CMS/TWikiLUM

  • 2016APV: 19.52 fb^-1
  • 2016: 16.81 fb^-1
    • /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/certification/Collisions16/13TeV/Legacy_2016/Cert_271036-284044_13TeV_Legacy2016_Collisions16_JSON.txt
  • 2017: 41.48 fb^-1
    • /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/certification/Collisions17/13TeV/Legacy_2017/Cert_294927-306462_13TeV_UL2017_Collisions17_GoldenJSON.txt
  • 2018: 59.83 fb^-1
    • /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/certification/Collisions18/13TeV/Legacy_2018/Cert_314472-325175_13TeV_Legacy2018_Collisions18_JSON.txt

Simulation

Samples and Generator Parameters

Sample csv files containing DAS strings, xrootd redirectors, and cross sections are stored the samples directory.

Sample Weights

Each sample is weighted by the data-MC luminosity ratio to normalize the expected MC contribution to the observed data contribution.

DY+Jets Stitching

In order to increase the statistics in the phase space Z+jets events with 1-4 jets, we use dedicated exclusive DY+nJets, where n=1,2,3,4, samples in addition to an inclusive sample of DY+Jets with any number of jets. These events have to be carefully weighted to account for the fact that multiple samples contribute relevant events. The process of weighting these events is called MC stitching.