trackObservables

installation

Trainings will run on cmslpc, and you'll need an EOS area to store condor output. Here is the advised setup:

$ cd ~/nobackup/
$ mkdir Substructure-ROC
$ cd Substructure-ROC
$ cmsrel CMSSW_7_2_0
$ git clone https://github.com/nhanvtran/trackObservables.git
$ xrdfs root://cmseos.fnal.gov mkdir /store/user/${USER}/SubROC/training/weights

For generation of ntuples, you need to install FastJet and FastJet/contrib. Place the installation in processing as follows:

For FastJet:
$ cd processing/
$ mkdir fastjet
$ cd fastjet
$ curl -O http://fastjet.fr/repo/fastjet-3.1.3.tar.gz 
$ tar zxvf fastjet-3.1.3.tar.gz
$ cd fastjet-3.1.3/
$ ./configure --prefix=$PWD/../fastjet-install
$ make && make check && make install

For FastJet/contrib:
$ cd ..
$ svn checkout http://fastjet.hepforge.org/svn/contrib/trunk fjcontrib
$ cd fjcontrib/
$ scripts/update-contribs.sh 
$ scripts/update-contribs.sh EnergyCorrelator 1.2.0-rc1
$ ./configure --fastjet-config=$PWD/../fastjet-install/bin/fastjet-config
$ make && make check && make install

running

Running sh steerScript.sh will show a number of commands. In the order used for the Track Observables analysis:

Usage: sh steerScript.sh <TASK> <OPTION>
  * MAKE      - build processing/anaSubstructure.cpp executable             
  * JOBS      - launch condor jobs for ntuple generation 
  * MERGE     - merge outputs from EOS after jobs have finished
  * PLOT      - plot output from merged jobs 
  * WWW       - send plots to CERNWEB: <OPTION> changes location
  * TRAIN     - run BDT trainings (settings inside file)
  * BDTS_MOVE - move BDT trainings from EOS after jobs have finished
  * BDTS_PLOT - plot BDT results
  * BDTS_WWW  - send BDT plots to CERNWEB: <OPTION> changes location

analysis

The main body of the BDT training. Scripts to configure and launch TMVA.

You can run trainings via condor using launchJobs.py, for example:

python analysis/launchJobs.py -b --doTraining --makeROCs --treeName t_allpar

Input files are assumed to have the form <prefix>-<type>-<ptfix>-<postfix>.root. Unless specified otherwise, the defaults for these tags are:

prefix : processed-pythia82-lhc13, change with --prefix
type : no default, but can be e.g. qq, tt, WW, ZZ, gg
ptfix : pt1, change with --ptfix
postfix : 50k, change with --postfix

Individual trainings can be run locally via analysis.py. The following will produce one of the BDTs produced in the condor implementation:

nohup python trackObservables/analysis/analysis.py -b \
--sampleDir /uscms_data/d2/ntran/physics/FCC/trackObservablesStudy/trackObservables/processing/prod-Jun14  \
--weightDir ./weights  \
--plotDir   ./plots    \
--sigTag    WW-pt1     \
--bkgTag    ZZ-pt1     \
--inputs    "j_tau21_b1[0];j_tau21_b2[0];j_c1_b0[0];j_c1_b1[0];j_c1_b2[0];j_c2_b1[0];j_c2_b2[0];j_d2_b1[0];j_d2_b2[0];j_mass_trim[0]*j_ptfrac[0];j_mass_mmdt[0]*j_ptfrac[0];j_mass_prun[0]*j_ptfrac[0];j_mass_sdb2[0]*j_ptfrac[0];j_mass_sdm1[0]*j_ptfrac[0];j_mass[0]*j_ptfrac[0]" \
--treeName t_allpar    \
--doTraining > ./output/WW-ZZ-pt1-tracks.txt

To reiterate, --prefix and --postfix will specify the input root files to pull from --sampleDir.

To plot ROCs from trainings:

$ root -l -b
[0] .L analysis/plotROC.C
[1] plotROC(input1, label1, input2, label2, ..., inputN, labelN, outputName) 
[2] .q

To get separation information from BDT trainings (condor only), run python getSeparationTXT.py from the directory where you launched the condor jobs. This will output .txt files with the desired tables.

plotting

Utilities to get formatted control plots from ntuples.

processing

To run the ntuplizer locally, look at steerScript.sh ANATEST, or:

$ cd processing
$ make
$ ./anaSubstructure pythia82-fcc100-gg-pt5-50k \ # LHE filehandle to look at
                    <location of LHE files> \ 
                    0 \   # index of first event to process
                    0 \   # index of last event to process 
                    rth   # options of detector smearing in random order (r: hcal resolution, t: trackpt-inefficiency, h: hcal-granularity, p: neutral pileup, pi: pileup+PUPPI, e: ecal resolution, u: tracking-efficiency, s: trackDr-inefficiency)

To run the ntuplizer via Condor (LPC):

$ cd processing
$ make
$ python SubmitCondor.py --indir <location of LHE files>                      \
                         --outdir <where to put all outputs: .condor & .root> \
                         --maxEvents <number of events to process>            \
                         --evPerJob  <number of events for each condor job>   \
                         --anaSubLoc <location of the anaSubstructure executable, by default in processing/> \
                         --fastJetLoc <location of the fastjet-install/ directory, by default in processing/fastjet/>

Note that you shouldn't have to change the last two options if you've followed the installation instructions above and are running these commands from processing/.