Core Minutes 10/18/2005  

Present:    Joanne Bogart, Philippe Bruel, Toby Burnett, David Chamont, Jim Chiang, Seth Digel, Richard Dubois, Warren Focke, Riccardo Giannitrapani, Berrie Giebels, Tom Glanzman, Navid Golpayegani, Tony Johnson, Heather Kelly, Michael Kuss, David Landriu, Chuck Patterson, Igor Pavlin, James Peachey, Leon Rochester, Tom Stephens, Johann Cohen-Tanugi, Tracy Usher

• Announcements: (Richard)
  • Farewell, Matt. He will be leaving us at the end of the month to pursue other opportunities. This will naturally require some regrouping on our part, and a search for a replacement.
  • We believe the 16th tower goes in today!
• DC2 news:
  • Event production (Richard, Tom G.) A 500 million event run with GlastRelease v7r2 is now complete; now aiming for final 1 billion event run. (Toby) Time for backgrounds is not distributed evenly. (Richard) We need to change from 30-second spread to 40-second spread; that will fix it. Tom has been adding diagnostics so that, if something does go wrong in these large Pipeline runs, we'll have some tools to find it quickly. He is also working on after-task analysis of Pipeline jobs.
  • Tracker (Leon) Two projects are in the works to help find good events. 1) Surplus hit ratio: In the past information in Tracker surplus hit ratio has been watered down by the approximately 40 noise hits per event. Bill suggests only looking at those layers with both X and Y hits belonging to the track, which reduces noise hits to only about 4 per event. Also we are going to replace the cylindrical volume around the track (or track vertex) with a cone, opening angle dependent on energy. To finish this off, need a way to make and adequate estimate of energy. 2) Counting hits above a track: these indicate a messy interaction.
  • Active Distance Calculation (Heather) is working on an improvement to the active distance calculation suggested by Bill; she hopes to have it done this afternoon. It concerns tracks which don't pierce any tiles, but are nearly parallel to a tile plane. In such cases the current method leads to a very large active distance so the variable ends up getting the default value. The new approach, 3-d in nature rather than being restricted to considering distances in the plane of the tile, is closer to the original intent.
  • Classification trees (Toby) In order to duplicate some of Bill's classification tree work in Gleam, would be nice to add some ntuple variables which are simple arithmetic combinations of existing variables. Otherwise Toby has to compute them elsewhere, complicating the general scheme.

    (Tracy) At Richard's suggestion he has been investigating Orange, a free, open source data mining package. The interface is similar to Insightful Miner. It's written in C++; functions can be accessed via Python scripts or through a GUI. We can hope that, since it is open source, those features it lacks will be added in time by the community using it, maybe even by us. It can't handle large ntuple files: it choked on one and Tracy had to cut it down to 5000 events. Originally Insightful Miner had a similar problem, but a work-around was found; perhaps one exists for Orange also. Tracy was able to build from sources on Windows. Would be useful to have Bill try it out and report impressions.

  • Possible CalDigi problem (Philippe) See this thread in the CalSoft mailing list for evidence of a glitch in the energy distribution in the vicinity of 1050 MeV, around the crossover from lex1 to hex8. (Berrie) System Tests don't have the right set of histograms to detect this kind of problem.
  • Too much randomness (Michael) He found small discrepancies — in the 4th or 5th decimal place — in particles going into G4, so he doesn't believe Geant is to blame. He found some uninitialized variables and memory leaks in our stuff, but they appear to be unrelated to reproducibility problems.
  • Crashes (Richard) There is evidence in core dumps of a problem in CAL-land, perhaps having to do with spline calculations.
• MRvcmt: (Riccardo) He's working on better handling of container packages; hopes to be done with this next week. Also, at Toby's request, he recently added the ability to invoke MRvcmt, passing CMTPATH file as a parameter, like this:

  ruby vcmt.rb -f e:\\Ric\\file.cmtpaths

  This way you can have two instances of MRvcmt with different CMTPATH running concurrently.
• Extra-curricular: Make it a musical week-end with Leon and Joanne .

previous

minutes index

next

J. Bogart, Last Modified: 01-Jun-2010 15:47:01 -0700