Tracker Recon Restructuring Report
Analysis Group Meeting
September 13, 2004
 

• Goals of re-structuring TkrRecon
  • TDS Class Restructuring
    • Restructuring eliminates the TkrPatCand and TkrPatCandHit classes
      • Output of Pattern Recognition/Input to Track Fit
      • Replaced with new Track class
      • Not used outside TkrRecon anyway...
    • Modifications to TkrCluster
      • Primary change is to include idents::TkrId instead of the old tower, layer, view information previously stored there.
        • Old "layer" in "recon order" anyway...
      • TkrClusterCol class (with all of its methods) to become purely a Gaudi ObjectContainer
        • Most access methods will be moved to TkrQueryClustersTool in TkrUtil
    • New Track class
      • TkrTrack will be the TDS track object
      • TkrTrackHit will be the TDS object containing the information for each hit on the track
      • TkrTrackParams will replace the TkrFitHit/TkrFitPar/TkrFitMatrix classes which contain the track parameters (of each type) at each hit
        • Biggest difference here is that TkrTrackParams is purely a data storage class and all the "matrix/vector math" operations are moved to outside helper classes
      • TkrTrackCol will be an ObjectVector containing the TktTrack objects
      • In principle, the new organization is Gaudi Converter compliant
    • TkrVertex will also be modified to follow a new vertexing plan from Bill Atwood
    • All of these changes in the Event Package (at the bottom of the food chain)
      • Changes here the most painful as they trigger rebuilds of nearly all Glast packages
         
  • TkrRecon Restructuring
    • Rationalize the code to make the current algorithms more easily understood
      • For example: stop looping over "layers" (which are really planes beginning at the top/front of the tracker
    • Move key pieces of code to the areas they are truly related to
      • For example: Kalman Filter hit finding code into pattern recognition
      • So, when Tracker experts make modifications they modify the right code!
    • Improve the track fitter to allow for easily changing input paramters
      • Second pass recon, test modes, etc.
      • Remove important calculations from the TDS classes
        • energy loss
        • chi-square
    • Keep open the possibility of providing "hooks" for alternative recon modes
    • TkrRecon a container package and make key pieces sub-packages?
      • Have looked at but appears will be quite a bit of work
         
• Current Status
  • Status of code retrofit of draft new classes into Recon
    • TkrCluster integration:
      • First draft (which includes idents::TkrId) integrated into code
      • TkrClusterCol is now a pure ObjectVector (as opposed to a more complex container class)
    • TkrTrack integration:
      • Integrated into Combo and Monte Carlo Pat Recs
        • Also retrofit to other two (unused) pat recs but not tested
      • Runs through the "generic" track fit
        • "Old" track fits used in combo pat rec in previous form
      • Runs through the TkrRecon vertexing
      • Further upstream:
        • All AnalysisNtuple tools that use tracks have been upgraded
        • AcdRecon converted to use TkrTracks
        • Have not addressed Root IO yet (but... is straightforward)
    • TkrVertex:
      • Current version modified to work with TkrTrack classes
      • New version still in draft stage
  • Overall
    • Have generated and reconstructed events through the full chain (except outputting the full root trees)
    • All modified code placed into cvs in the users/TkrGroup area
      • Builds against the head of GlastRelease (to stay as current as possible)
      • Would like to branch the various packages and introduce the code into the main cvs areas to ease future integration
    • Have a fully built version of Gleam running this code on Glast-TS
    • Warning: Have NOT updated the gui display code
      • currently only supporting the Fred display code!
      • A good time to make the switch?
         
• Still quite a bit to do
  • Restructured TDS classes integrated (as much as possible) into current code is first step
  • Now need to evaluate these classes
  • Begin restructuring TkrRecon itself in order to maximize use of the new classes
     
• Current Schedule:
  • Upcoming I&T Workshop will slow us down for next two weeks
  • Original plan was:
    • Finalize draft classes (for integration into code): end of July
      • Was mid-August
    • Integration into existing code and released for Tracker group use: end of August
      • Was September 10
    • First go round of restructured TkrRecon code in September
      • Goal was to "push the button" on September 19
  • Updated plan (including 2 week slide for I&T)
    • September: Finalize structure of TDS classes, plan for restructuring TkrRecon/TkrUtil (adding utility classes for accessing info from TkrTrack/TkrClusters)
    • October: Implement these modifications
      • Goal: push the button on October 22
      • Use to the end of October to "fix" problems
    • Re-integrate into the main code branch?
   

T.Usher Last Modified: 09/13/2004 08:36