Skip to content

Latest commit

 

History

History
355 lines (267 loc) · 16.4 KB

README.md

File metadata and controls

355 lines (267 loc) · 16.4 KB
Raw

reco-prodtools

Home of the tools to produce samples for HGCAL reconstruction studies

Getting started

Set up CMSSW according to github.com/CMS-HGCAL/reco-ntuples.

Then anywhere,

git clone [email protected]:CMS-HGCAL/reco-prodtools.git reco_prodtools
cd reco_prodtools/templates/python
./produceSkeletons_D41_NoSmear_noPU.sh
cd ../../..
scram b
cd reco_prodtools/
python SubmitHGCalPGun.py --nevts 2 --evtsperjob 1 --queue 1nh --partID 13 --thresholdMin 35 --thresholdMax 35 --gunType E --tag test_${USER}

Mind that the directory reco_prodtools needs to have the underscore (instead of a hyphen).

Available configurations

For details on the geometry, please see Configuration/Geometry/README.md. For details on the pileup scenario, please see Configuration/StandardSequences/python/Mixing.py. Also note that running the GSD step for the pileup scenario requires a valid voms proxy (voms-proxy-init [-voms cms]).

Snippet Era Geometry Beamspot PU
produceSkeletons_D41_NoSmear_noPU.sh Phase2C8 D41 NoSmear none
produceSkeletons_D41_NoSmear_PU_AVE_200_BX_25ns.sh Phase2C8 D41 NoSmear AVE_200_BX_25ns
produceSkeletons_D41_VtxSmearedHLLHC_noPU.sh Phase2C8 D41 VtxSmearedHLLHC none
produceSkeletons_D44Nose_NoSmear_PU_AVE_200_BX_25ns.sh Phase2C6_timing_layer_bar D44 NoSmear AVE_200_BX_25ns

Whenever you would like to change configuration, change to the reco_prodtools/templates/python directory and execute the corresponding script. Then make sure to run scram b.

The processing of TICL iterations is added into the RECO fragment by default. To disable that, add no-ticl to the skeleton creation script, e.g.:

./produceSkeletons_D41_NoSmear_noPU.sh no-ticl

Other options like changing the pileup source, calling --customise on cmsDriver or changing geometry can be passed with the following script

./produceSkeletons_D41_NoSmear_PU_AVE_200_BX_25ns_aged.sh \
        pileup_input=/eos/cms//store/group/dpg_hgcal/comm_hgcal/deguio/pu_library/V11/GEN-SIM \
        custom="--customise SLHCUpgradeSimulations/Configuration/aging.agedHGCal" \
        geometry=Extended2026D46

The following script produces the configuration for producing samples with the HFNose sub-detector.

./produceSkeletons_D44Nose_NoSmear_PU_AVE_200_BX_25ns.sh

In order to store HF+Nose hits (and switch off HGCAL hits) in the NTUP step, you need to set the option detector = cms.string("HFNose") in the templates/partGun_NTUP_template.py configuration file.

Available guns and processes

The production tools allow you to generate a plethora of processes (links lead to implementation in CMSSW):

Details

To produce NEVENTS GEN-SIM-DIGI events with NPART sets of particles (per event) of type PART_PDGID and in the p_T range from PTMIN to PTMAX, one should run:

  python SubmitHGCalPGun.py
  --datTier GSD
  --nevts NEVENTS
  --evtsperjob NPERJOB
  --queue QUEUENAME
  --partID PART_PDGID
  --nPart NPART
  --thresholdMin PTMIN
  --thresholdMax PTMAX
  --gunType GUNTYPE
  [--gunMode GUNMODE]
  [--local]
  --tag MYTAG

Here, one can produce a custom set of particles by providing PART_PDGID as a set of comma-separated single PDG IDs. To simulate the decay of unstable particles, e.g. quarks, gluons or taus, an alternative particle gun based on PYTHIA8 can be used by setting --gunMode pythia8. The Pythia8 gun can also be either a 'Pt' or 'E' gun. Mind that if --thresholdMax PTMAX is not provided (or is negative), it will be set to the value of thresholdMin (and therefore produce events with a fixed p_T/energy).

To produce NEVENTS GEN-SIM-DIGI events with pair of particles within given angular distance ΔR(η,φ) (per event), where the first particle is of type PART_PDGID and in the p_T range from PTMIN to PTMAX, and the second one is of type INCONE_PART_PDGID and at distance from DRMIN to DRMAX and with p_T in range from PTRATIO_MIN to PTRATIO_MAX relative to the first particle, one should run:

  python SubmitHGCalPGun.py
  --datTier GSD
  --nevts NEVENTS
  --evtsperjob NPERJOB
  --queue QUEUENAME
  --partID PART_PDGID
  --nPart 1
  --thresholdMin PTMIN
  --thresholdMax PTMAX
  --gunType Pt
  [--gunMode GUNMODE]
  --InConeID INCONE_PART_PDGID
  --MinDeltaR DRMIN
  --MaxDeltaR DRMAX
  --MinMomRatio PTRATIO_MIN
  --MaxMomRatio PTRATIO_MAX
  [--local]
  --tag MYTAG

One should also note that for the genertion of pairs of particles within a given cone, one has to use the "Pt" gun. Also note that it is currently not possible to generate pairs within a cone using the PYTHIA8-based gun.

The script will create a directory called partGun_[MYTAG]_[DATE] locally or on the CMG EOS area (see options), and submit jobs to queue QUEUENAME with NPERJOB events per job, NEVENTS in total. The batch stdout/stderr files and .cfg files used to run are stored locally in partGun_[MYTAG]_[DATE], while the resulting files partGun_*_GSD_{i}.root are stored in partGun_[MYTAG]_[DATE] either locally or in /eos/cms/store/cmst3/group/hgcal/CMG_studies/Production/.

Rule of thumb for GEN-SIM-DIGI: 4 events per 1nh:

Close-by gun

Another gun that could be used is --gunMode closeby, which is capable of creating several vertices. Mind that it is only available in CMSSW_10_6_0 or later. With this choice particles can be produced with random energy, R and Z in a specified range. When more than one particle are asked to be produced, then each particle will be created at a different vertex, equally spaced by Delta, the arc-distance between two consecutive vertices over the circle of radius R. Also, there is the --nopointing option: if used, particles will be produced parallel to the beamline; by default, the particles will be pointing to (0,0,0). Furthermore, there is the --overlapping option that if used then particles will be generated in a window [phiMin,phiMax], [rMin,rMax], otherwise with a DeltaPhi=Delta/R. Another option is --randomShoot which if used will shoot a random number of particles. However, this option should be used alongside the --nRandomPart in order for the gun to know the upper limit on how many particles to shoot. The --nRandomPart option shouldn't be confused with the size of the --partID option, since with --partID we are setting the particles we are interesting in producing, while with --nRandomPart we are randomly choosing the number we want to shoot out of those --partID ids. Apart from producing multiple particles, this gun could also produce a single particle wherever the user wishes, having always the nice feature of assigning to the vertex the time required to travel from (0,0,0) to the desired location. This could be useful e.g. when someone wants to shoot straight to the scintillator part. Keep in mind that in this case there is no sense of neither adding the antiparticle nor adding the --randomShoot option. As an example, the command below will produce NEVENTS GEN-SIM-DIGI events with NPART sets of particles (per event) of type PART_PDGID in the energy range from EMIN to EMAX (Pt option not available), radius range from RMIN to RMAX, z position from ZMIN to ZMAX, parallel to the beamline, with a distance between the particles vertices of deltaPhi = DELTA/R.

  python SubmitHGCalPGun.py
  --datTier GSD
  --nevts NEVENTS
  --evtsperjob NPERJOB
  --queue QUEUENAME
  --partID PART_PDGID
  --nPart NPART
  --thresholdMin EMIN
  --thresholdMax EMAX
  --rMin RMIN
  --rMax RMAX
  --zMin ZMIN
  --zMax ZMAX
  --Delta DELTA
  --pointing
  --etaMin ETAMIN
  --etaMax ETAMAX
  --gunType E
  --gunMode closeby
  --tag MYTAG

Physics processes

The physics processes listed above can be generated as follows using gunMode set to physproc and the gunType to the process of your choice:

  python SubmitHGCalPGun.py
  --datTier GSD
  --nevts NEVENTS
  --evtsperjob NPERJOB
  --queue QUEUENAME
  --tag MYTAG
  --gunMode physproc
  --gunType wqq

In addition to setting the physics process only, you can add a filter path to have the events enriched by a minimum number of jets with a pT above a certain threshold within the HGCal coverage (1.479 < |eta| > 3.0). To do so, extend the gunType:

--gunType wqq:ak8GenJetsNoNu:100:2

which in this case means at least two jets with pT > 100 GeV in the ak8GenJetsNoNu collection.

A more efficient way to apply phase space cuts is to directly configure Pythia8 via process parameters. Currently supported cuts:

  • PhaseSpace:pTHatMin: set via --thresholdMin, ignored when negative.
  • PhaseSpace:pTHatMax: set via --thresholdMax, ignored when negative.

RECO step

To run RECO stage on the produced GEN-SIM-DIGI samples, stored under partGun_[MYTAG]_[DATE] (locally or on the CMG EOS area), one should run:

python SubmitHGCalPGun.py \
  --datTier RECO \
  --evtsperjob NPERJOB \
  --queue QUEUENAME \
  --inDir partGun_[MYTAG]_[DATE] \
  [--local] \
  [--tag NEWTAG] \
  --keepDQMfile

The script will get the list of GEN-SIM-DIGI files from the directory partGun_[MYTAG]_[DATE]/GSD (locally or on the CMG EOS area), and submit jobs to queue QUEUENAME (if possible with NPERJOB events per job). The batch stdout/stderr files and .cfg files are stored locally in partGun_[MYTAG]_[DATE], while the resulting files partGun_*_RECO_{i}.root are stored in partGun_[MYTAG]_[DATE]_[NEWTAG]/RECO (_NEWTAG is only added if --tag is used) either locally or in /eos/cms/store/cmst3/group/hgcal/CMG_studies/Production/. In case the --keepDQMfile option is used, the resulting partGun_*_DQM_{i}.root files will also be stored in partGun_[MYTAG]_[DATE]_[NEWTAG]/DQM locally or in /eos/cms/store/cmst3/group/hgcal/CMG_studies/Production/.

Rule of thumb for RECO: 10 events per 1nh:

  • 50 events should be possible to finish in queue 8nh.
  • Ditto, 1000 events in 1nd.

For more info on available options type python SubmitHGCalPGun.py --help

NTUP step

This needs the ntupliser to be installed as in github.com/CMS-HGCAL/reco-ntuples

To run NTUP stage on the produced RECO samples, stored under partGun_[MYTAG]_[DATE] (locally or on the CMG EOS area), one should run:

python SubmitHGCalPGun.py \
  --datTier NTUP \
  --evtsperjob NPERJOB \
  --queue QUEUENAME \
  --inDir partGun_[MYTAG]_[DATE] \
  [--multiClusterTag hgcalLayerClusters] \
  [--noReClust] \
  [--local] \
  [--tag NEWTAG]

Mind that the multiClusterTag option needs to be provided for RECO files created before CMSSW_10_3_X.

ALL steps

Running the 3 steps together in one single submission is possible by specifying the --datTier ALL option. Please note that if this option is activated only the ```NTUP''' is stored in the output directory.

RelVal

It can also run directly on RelVal using the same NTUP cfg. It runs das_client internally so you need to have a valid proxy (i.e. run voms-proxy-init before). The output goes in an area (eos/local) named after the RelVal dataset with all '/' replaced by underscores.

Typical usage:

python SubmitHGCalPGun.py \
  --datTier NTUP \
  --queue QUEUENAME \
  [--local] \
  -r /RelValSingleMu..etc.

Notes

Starting from CMSSW_10_4_0_pre3 (cms-sw/cmssw#25208), the handling of the calibration weights has been rewritten to be more generic and play nicely with eras. PR #54 took care of these changes.

Updating and adding configurations

To update and add new configurations, you need to pick a geometry as listed in Configuration/Geometry/README.md, and then run the following command grepping for your desired geometry (here D41) and usually also a process of interest (here SinglePiPt):

runTheMatrix.py -w upgrade -n | grep D41 | grep SinglePiPt

This will result in output similar to the following:

20488.0 SinglePiPt25Eta1p7_2p7_2026D41_GenSimHLBeamSpotFull+DigiFullTrigger_2026D41+RecoFullGlobal_2026D41+HARVESTFullGlobal_2026D41
20688.0 SinglePiPt25Eta1p7_2p7_2026D41PU_GenSimHLBeamSpotFull+DigiFullTriggerPU_2026D41PU+RecoFullGlobalPU_2026D41PU+HARVESTFullGlobalPU_2026D41PU

In the above case, there are two workflows listed, one with and one without PU. Let's pick the one with PU, 20688.0, and get the configs:

runTheMatrix.py -w upgrade -l 20688.0 --command="--no_exec" --dryRun

This will run a while and create a new directory that contains the configs (e.g. 20688.0_SinglePiPt25Eta1p7_2p7_2026D41PU_GenSimHLBeamSpotFull+DigiFullTriggerPU_2026D41PU+RecoFullGlobalPU_2026D41PU+HARVESTFullGlobalPU_2026D41PU). Further, several cmsDriver.py scripts will be printed out to the screen. These are the ones that need to be adjusted and put into the corresponding shell scripts (have a look at the existing shell scripts themselves).

Some general guidelines:

  • For all commands remove --filein and --fileout options.
  • Add python_filename option

The first command combines step1 and step2 (GSD):

  • mix in pileup (for 200 PU use e.g. --pileup AVE_200_BX_25ns)
  • run up to DIGI...HLT:@fake2

The following changes are implemented on top/need to be adjusted:

  • --beamspot HLLHC14TeV ➜ --beamspot NoSmear (if you don't want a smeared beamspot)
  • --eventcontent FEVTDEBUG ➜ --eventcontent FEVTDEBUGHLT (since step2 has that)

The second command is step3 removing overlap with step2 (RECO):

  • remove pileup part
  • also remove MINIAODSIM, PAT
  • remove from VALIDATION@miniAODValidation
  • remove from DQM:@miniAODDQM

The third command is a copy of the second only re-running RECO (for NTUP):

  • remove DQM
  • add processName=NTUP option

For more details see the configuration files listed above.

Contributing

We use the fork and pull model:

Fork this repository.

If you haven't done so yet, clone this repository:

git clone [email protected]:CMS-HGCAL/reco-prodtools.git reco_prodtools

Add your fork of the repository as remote:

git remote add mine [email protected]:`git config user.github`/reco-prodtools.git
git checkout -b ${USER}_feature_branch origin/master

Work on your feature, add, commit, etc.

git fetch origin
git rebase origin/master
git push mine feature_branch

Make a pull request on github.