Data Quality Check Procedures and Tools for Monte Carlo Production

1
Data Quality Check Procedures and Tools for Monte
Carlo Production

• Eric van Herwijnen
• Miriam Gandelman
• September 17th 2003

2
Contents

• Objectives
• Quality Check Procedures
• CMT packages
• Reference Data (Boole, Brunel, Gauss, SICBMC)
• Tools
• Integration into DIRAC
• Conclusions

3
Objectives

• Procedure for checking MC data
• Tools for local managers to check remotely
• Check log files for small nb of important
  quantities
• Integrate into DIRAC
• Tools for sw managers to check consistency
  between different versions
• Tools to analyze histograms of large samples
  (50k)

4
Quality Check Procedures

• New versions of Boole, Brunel
• Quality quantities defined for Trigger, Velo
  (Boole) and Inner/Outer Tracker and Rich (Brunel)
• Use reference tables of quantities from previous
  production
• Tools to compare quantities from a log file (in
  production) with the reference values -gt quality
  report web page
• Tools to compare histograms with reference sets
• Quality of production data
• Make new reference tables when data is okd by
  physics coordinator
• Quality report Brunel_QA_xxx_yyy.html (xxx
  production number, yyy jobnumber)
• Deviations gt 3sigma from reference values are
  printed in red

5
CMT packages

• SICB/SICBMCquality v1r0. Histograms made with
  v260r1,2,3.
• Quality/Boole v1r0. Tables and histograms made
  with Brunel v18r1 and Boole v1r0.
• Quality/Brunel v1r3. Tables and histograms made
  with Brunel v18r1 and v20r0p1.
• Quality/Tools v1r1. Python (analyse log files),
  c (root for histograms)
• Reference histograms are stored also at
• http//lhcb-wdqa.web.cern.ch/lhcb-wdqa/vol11/packa
  gename/packageversion/evttype/index.html

6
Reference Data (Boole)

• Boole v1r0
• Tables
• Trigger L0 Acceptance, L1 Efficiency
• Reference sets made with Brunel v18r1
• Histograms
• Velo (MCVeloHits, VeloClusters), IT
  (MCITDepositCheck, MCITDigitChecker,
  ITDigitChecker), Rich (DIGI, occupancies)
• Histograms made with Boole v1r0

7
Reference Data (Brunel)

• Brunel v18r1
• Tables
• Overall tracking efficiency, number of
  reconstructed Ks, IT efficiencies, Pi/K
  efficiencies and misID rates
• Tables made with Brunel v18r1
• Histograms
• ITClusterChecker, RICH PIDs, OTClusterChecker,
  OTClusterMonitor
• Histograms made with Brunel v20r0p1

8
Reference Data (Gauss)

• No histograms printed by Gauss
• Standard Gaudi algorithms could access
  subdetector information (nb of hits) and
  generator information

9
Reference data (SICBMC)

• SICBMC v260r1,2,3
• Tables
• Efficiency of acceptance cuts
• Histograms
• Outer tracker, Velo, Rich, Ecal, Hcal, Muon (nb
  of hits) and Generator (multiplicities,
  production/decay points, pileups, primary z vtx)

10
Tools

• Log file analysis tools
• Use python 2.2
• Require logfiles to be accessible from
  bookkeeping database
• Reference sets are made by hand, comparisons
  automatically done in production
• Histogram analysis tools
• Use the wdqa package (add hbook files, convert to
  root, create gif images, web pages)
• Use Kolmogorov test for comparisons

11
Integration into DIRAC

• Quality tables packaged with Dirac
• Tools called from scripts via environment
  variables
• Output of tools are html pages, put in the same
  web location as the log file

12
Conclusions

• Need more experience of tools in production
  initial tests in Lyon succesful
• Need quality information from missing
  subdetectors
• Need to write Gaudi algorithms for Gauss tables
  and histograms
• More details in note LHCb 2003-122
• http//lhcb-comp.web.cern.ch/lhcb-comp/dataquality
  /dqnote.pdf