Skip to content

Latest commit

 

History

History
81 lines (55 loc) · 5.03 KB

README.md

File metadata and controls

81 lines (55 loc) · 5.03 KB
Raw

UWAnalysis, ZZ version

Quickstart recipe

scram pro -n zz535 CMSSW CMSSW_5_3_5
cd zz535/src                                                                                                                                                                                                    
cmsenv
git clone https://github.com/iross/UWAnalysis.git
sh UWAnalysis/recipe.sh
scram b -j 4

Quick Explanation

Each analysis is driven by a python file in UWAnalysis/Configuration/python. This file uses the CutSequenceProducer python class in tools sub directory to create a series of cuts. The user can plug an ntuple plotter in any step of the selection (i.e preselection for background estimates or final selection for shapes). The file tools/analysisTools.py has the core configuration which is common for the analysis. After that the UWAnalysis/CRAB directory has the different final states. There is a file for MC and a file for data. If you look in the py file you will see that it loads the sequence defined in analysisTools. There are also other potential switches to add systematics (shift and clone selection sequence) or add ntuples for each systematics or at preselection selection level. The user can easilly change those.

The ntuplization is absolutely done with python and is controlled by the file UWAnalysis/Configuration/python/tools/zzNtupleTools.py.

Development is being done on the way the ntuples are structured. As of 2012-11-14, the default mode on the master branch is to save one candidate combination per event. In the event that multiple candidates pass the selection, the chosen one is defined as the one at the head of the grand list of candidates (different methods of sorting can be applied). (NOTE: this isn't entirely stable.. sometimes I check in the 'multiple candidate' mode. Be sure to check the LLLL and LLLL-MC.py to see which it is (if leadingOnly is True in the ntuple fillers, then it's only a single candidate stored per event))

Submitting Jobs

Job submission is handled by local (to UW) scripts. Job submission and retrieval scripts are produced via the CRAB/[ZH,LLLL]/submitJobs.py and CRAB/[ZH,LLLL]/mergeJobs.py scripts. The datasets.json file is used to define the datasets of interest (and some of their properties).

To create the jobs:

python submitJobs.py --tag=[identifying tag]

By default, this will create job submissions for ALL datasets in datasets.json. To run over a subset, pass a search term (or terms) via the --samples option:

python submitJobs.py --tag=data2012_testRun --samples "Double*2012*" "MuE*2012*"

Then submit the jobs by running the freshly created shell script:

sh submitJobs.sh

When jobs are complete, running similar mergeJobs commands will create farmout jobs to merge subsets of the data, which must then be added together by hand to create the final ntuple.

python mergeJobs.py --tag=data2012_testRun --samples "Double*2012*" "MuE*2012*"
sh mergeJobs.sh
#once these jobs are done, you'll need to go to /scratch/$USER/ and hadd the merged output files

BG Estimation

The current analysis flow for putting together a complete BG estimate is:

  1. Run jobs

    per usual, being sure to use "leadingOnly=False" in the ntuple fillers (e.g. https://github.com/iross/UWAnalysis/blob/master/CRAB/LLLL/LLLL.py#L80) and a loose config as the runtime selection (e.g. https://github.com/iross/UWAnalysis/blob/master/CRAB/LLLL/LLLL.py#L57)

  2. Merge jobs.

  3. Post-processing.

    Because each event can produce multple candidates for each event, we must apply some post-processing.

     cd analysisScripts
 python makeAnalysisTrees.py --file=[inputFilepath].root --out=[outputFilepath].root

    By default, this creates fully selected trees, BG estimate trees (Z+1l for measurements, Z+1P1F, Z+2F regions), same-sign BG estimate trees for checks, and some summed trees. Arbitration is applied, ensuring that the base selection trees only have one candidate per event. The resulting BG estimation treest can have multiple candidates per event (each candidate has the potential to fake a signal-like event)

  4. Plotting.

    analysisPlots.py in https://github.com/iross/analysisScripts/ has some pretty ugly plot dump code in place, which measures and applies the relevant fakerates. Alternatively, you can use the helper functions in analysisScripts/plotHelpers.py, which return fakerate-scaled histograms from all BG regions:

     from plotHelpers import *
 from ROOT import TFile
 f=TFile("whatever.root")
 applyFakes(f,extra="&&z1Mass>81&&z1Mass<101",lowZ1=True)

PatTuples

Default running mode is on the common UW pattuples. Samples are being produced centrally.

For information on creating/using the pattuples, see relevant sections of http://final-state-analysis.readthedocs.org/en/latest/index.html or ask one of us..

Previous Documentation

https://twiki.cern.ch/twiki/bin/view/Main/UWDiTau