Muon PID with a uboost BDT (in ROOT 5). Also include code for PID efficiency studies. Below we list some related links:
- A list of centrally produced calibration data
- WG prodcution twiki
- PIDCalib package twiki
- PIDCalib sample modes and cuts option file
make AddUBDTBranchRun2 # compile it first
./bin/AddUBDTBranchRun2 -i <input_ntuple> -o <output_ntuple> -p <muon_branch_name, like "probe"> -x <UBDT_xml> -t <treename>,[<treename> ... ]
Retraining is broken, and there's no plan to fix. The code is preserved in test
for archival purposes.
The sample is produced centrally with request ID 69972.
They are available at this eos location:
/eos/lhcb/grid/prod/lhcb/LHCb/Collision16/PIDCALIB.ROOT/00111665/0000
From DIRAC, this sample is produced with:
Conditions: Beam6500GeV-VeloClosed-MagUp type:
Processing pass: Real Data/Reco16/Turbo02a
Input file type: FULLTURBO.DST
DQ flag: OK
Input production: ALL
TCKs: ALL
Processing Pass: PIDCalibTuples9r3p1/PIDMerge05
Step 1 PIDCalibTuples9r3 2016 Jpsinopt(142199/PIDCalibTuples9r3p1) : Castelao-v3r4
System config: x86_64-centos7-gcc8-opt MC TCK:
Options: $PIDCALIBROOT/scriptsR2/makeTuples_pp_2016_reprocessing_Jpsinopt.py;$PIDCALIBROOT/scriptsR2/DataType-2016repro.py Options format: WGProd Multicore: N
DDDB: Condition DB: DQTag:
Extra: WG/PIDCalib.v9r3 Runtime projects:
Visible: Y Usable:Yes
Input file types: FULLTURBO.DST Output file types: DAVINCIHIST,PIDCALIB.ROOT
Step 2 PIDCalib-Ntuples-Merge(131852/PIDMerge05) : Noether-v1r4
System config: MC TCK:
Options: $APPCONFIGOPTS/DataQuality/DQMergeRun.py Options format: Multicore: N
DDDB: Condition DB: DQTag:
Extra: AppConfig.v3r332 Runtime projects:
Visible: Y Usable:Yes
Input file types: PIDCALIB.ROOT Output file types: PIDCALIB.ROOT
PIDCalib samples are generated with Castelao. For run 1 and 2,
Castelao/v3r4 is used, with CMTCONFIG=x86_64-centos7-gcc8-opt.
Follow these commands to setup a Castelao dev environment on lxplus:
lb-dev Castelao/v3r4
cd ./CastelaoDev_v3r4
git lb-use Castelao
git lb-checkout Castelao/run2-patches PIDCalib/PidCalibProduction
git lb-clone-pkg WG/PIDCalib -b v9r3
All required changes to Castelao are included in this repository. Copy all
contents in ./castelao to CastelaoDev_vXrY folder.
Finally, comment out the Input and DataType lines in the
makeTuples_pp_YYYY_reprocessing*.py files. Below is an example from
makeTuples_pp_YYYY_reprocessing_Jpsinopt.py file:
dv = DaVinci (
InputType = "MDST"
, Turbo = True
, RootInTES = "/Event/Turbo"
# , DataType = "2016" #added by me just for this test
, EvtMax = -1
, Lumi = True
, TupleFile = "pidcalib.root"
# , Input = ['root://eoslhcb.cern.ch//eos/lhcb/user/p/poluekt/PID/2016_TurCal_DST/00053197_00000610_2.fullturbo.dst']
)Now, to produce J/psi only samples, first, drop into a bash shell:
./run bash
In the bash shell, run:
gaudirun.py \
$PIDCALIBROOT/scriptsR2/makeTuples_pp_2016_reprocessing_Jpsinopt.py \
$PIDCALIBROOT/scriptsR2/DataType-2016repro.py
To produce everything else, drop in the same bash, then:
gaudirun.py \
$PIDCALIBROOT/scriptsR2/makeTuples_pp_2016_reprocessing.py \
$PIDCALIBROOT/scriptsR2/DataType-2016repro.py
First, spawn a container with an image built by us:
make docker-cl
Then follow the instructions in the previous section.
According to the ANA note, p.16, these are the input variables to the uBDT (These names are also used in the source):
TrackChi2PerDof <-> probe_Brunel_ANNTraining_TrackChi2PerDof
TrackNumDof <-> probe_Brunel_ANNTraining_TrackNumDof
TrackGhostProbability <-> probe_Brunel_ANNTraining_TrackGhostProb
TrackFitMatchChi2 <-> probe_Brunel_ANNTraining_TrackFitMatchChi2
TrackFitVeloChi2 <-> probe_Brunel_ANNTraining_TrackFitVeloChi2
TrackFitVeloNDoF <-> probe_Brunel_ANNTraining_TrackFitVeloNDoF
TrackFitTChi2 <-> probe_Brunel_ANNTraining_TrackFitTChi2
TrackFitTNDoF <-> probe_Brunel_ANNTraining_TrackFitTNDoF
####
RichUsedR1Gas <-> probe_Brunel_RICH1GasUsed
RichUsedR2Gas <-> probe_Brunel_RICH2GasUsed
RichAboveMuThres <-> probe_Brunel_RICHThresholdMu
RichAboveKaThres <-> probe_Brunel_RICHThresholdKa
RichDLLe <-> probe_Brunel_ANNTraining_RichDLLe
RichDLLmu <-> probe_Brunel_ANNTraining_RichDLLmu
RichDLLk <-> probe_Brunel_ANNTraining_RichDLLk
RichDLLp <-> probe_Brunel_ANNTraining_RichDLLp
RichDLLbt <-> probe_Brunel_ANNTraining_RichDLLbt
####
MuonBkgLL <-> probe_Brunel_ANNTraining_MuonLLBkg
MuonMuLL <-> probe_Brunel_ANNTraining_MuonLLMu
MuonNShared <-> probe_Brunel_ANNTraining_MuonNShared
InAccEcal <-> probe_Brunel_ANNTraining_InAccEcal
EcalPIDe <-> probe_Brunel_ANNTraining_EcalPIDe
EcalPIDmu <-> probe_Brunel_ANNTraining_EcalPIDmu
InAccHcal <-> probe_Brunel_ANNTraining_InAccHcal
HcalPIDe <-> probe_Brunel_ANNTraining_HcalPIDe
HcalPIDmu <-> probe_Brunel_ANNTraining_HcalPIDmu
InAccPrs <-> probe_Brunel_ANNTraining_InAccPrs
PrsPIDe <-> probe_Brunel_ANNTraining_PrsPIDe
InAccBrem <-> probe_Brunel_ANNTraining_InAccBrem
BremPIDe <-> probe_Brunel_ANNTraining_BremPIDe
VeloCharge <-> probe_Brunel_ANNTraining_VeloCharge
####
probe_isMuonTight <-> probe_Brunel_isMuonTight
####
TrackP <-> probe_Brunel_ANNTraining_TrackP
TrackPt <-> probe_Brunel_ANNTraining_TrackPtNote: Left column indicates the input branches for normal ntuples, right indicates equivalent branches in PIDCalib ntuples.
Naively, one would expect RichDLLe and probe_Brunel_RichDLLe are identical.
However, when comparing against each other:
idx RichDLLe probe_Brunel_RichDLLe
--- -------- ---------------------
12 -29.0931 -29.091299057006836
82 0.0 -1000.0
87 -19.7908 -19.787399291992188
96 -14.7653 -14.826299667358398
127 0.0 -1000.0
165 -2.0359 -1.315000057220459
200 0.0 -1000.0
229 -12.471 -12.465800285339355
230 -47.2728 -47.21480178833008
307 -55.0559 -55.055999755859375
341 -25.4451 -8.821700096130371
600 0.0 -1000.0
682 2.9152 2.915299892425537
701 0.0 -1000.0
702 -20.8758 -20.87540054321289
866 5.6615 5.662499904632568
916 -43.458 -43.38570022583008
1063 0.0987 0.13289999961853027
1123 0.0 -1000.0
1137 -2.8032 -2.803299903869629
1198 3.3484 3.355799913406372
1207 0.0 -1000.0
1247 -8.0879 -0.5562000274658203
1293 -0.1925 -0.1923999935388565
1311 0.0 -1000.0