Some tasks to improve STAR vertex finders including KFVF

Some tasks to improve STAR vertex finders including KFVF:

  • Confirm and investigate why we loose primary vertices reconstructed with KFVF in W samples (data/embedding) when comparing with PPVF/MinuitVF
  • The HF group uses KFVF to refit vertex seeds found by MinuitVF. Does it imply that the quality of the seeds found by KFVF is worse? In general, how does the seed quality (i.e. vertex finding efficiency) differ between KFVF and MinuitVF (and PPVF)? (see "Techinal details" below)
  • Update improve documentation of the code (doxygen)
  • We already have a tool to compare vertex finders. The work is ongoing and the direction where to move with it is clear. But contributions and ideas will certainly be welcomed
    • For example, need new feature to keep track of the time spent by the VFs reconstructing vertices
  • Technical details:
    • Comply with the existing generic vertex finder interface in StGenericVertexMaker/. Suggestons for additional abstract interfaces will be welcomed if there is a feeling we need to add one.
    • Designate/separate (when possible) individual stages in vertex finding/reconstruction: seeding, fitting, ranking, etc. By having clear reconstruction stages we can try to find the best combination.
      • Decouple seed finding (not always possible in adaptive algorithms but KFVF does have such stage)
      • Decouple ranking schemes from vertex finder implementation so we can apply them to vertices found by different finders
    • Separate primary and secondary vertex collections in KFVF
    • Use common options such as beamline, nobeamline, etc.
    • Move KFVF code from StiMaker/ to StGenericVertexMaker/ or similar place in order to simplify maintenace
    • Resurrect the study of TMVA ranking independent of any vertex finder