Minutes from 23 June 2009

2009/06/23 CSEP Meeting

Attendees: John Y., Masha L

1. V9.4.2 on csep-op
   • Re-processing of N, L and R tests for RELM groups for 2007
     1. 2007-08-02 and 2007-10-02 test dates failed due to NFS server problems
   • Stuck cseprandom processes during re-processing:
     1. Masha tried to write random numbers to the file to see if it fixes the problem
        • Ran for 48 hours without a single stalled process
        • Committed the work around to SVN
2. Matt reported slower processing of one-day New Zealand forecasts
   • Masha traced down to Matlab R2007b version vs. R2007a used by CSEP at SCEC
3. Tasks for CSEP V9.7 release:
   1. Release candidate is scheduled for Wednesday, June 24, 2009
      • To be installed on newly built csep-cert server
   2. John is working on adding daily results to the web viewer
   3. Matt asked for ability to re-process file-based forecasts in batch mode
      • DONE Trac ticket #159: will be in CSEP V9.7
   4. Masha updated web server tests with new S and M evaluation tests (per Trac ticket #155)
      • John and Masha are working on adding CSEPWebTest.py to nightly build
   5. DONE Peter Shebalin's EAST three-month alarm-based model with M>=4 for California
      • Peter validated forecast map published to intensity.usc.edu by csep-cert server
   6. DONE Yan's Global long-term and short-term models using CMT:
      1. Masha installed one-day and one-year models on csep-cert
         • Group agreed to have start date of 2010-01-01 for one-year model
         • One-day forecast will be tested starting 2009-07-01
         • Yan validated forecasts maps on intensity.usc.edu published by csep-cert
           1. Yan asked about status of Trac ticket #140: using 2 digits of scalar moment magnitude instead of 3 digits provided by CMT
           2. Masha tried the fix, but caused errors by the model
4. Jeremy provided a fix for N-test due to the comments by David Rhoades and Matt:
   • Will have delta1 and delta2 result variables vs. current one delta variable
   • Not necessary to have intermediate release because of the fix
   • May be a target for V9.10.0
   • Jeremy will confirm if daily N-test plot is affected by the change
   • Masha will start working on it once V9.7 release candidate is out
5. mini_CSEP distribution
   • Jeremy suggested to poll community for requirements
     1. N, L, R, M and S tests written in Python
     2. Conversion routines between XML and Matlab for the forecasts
     3. Plots, maps
   • We should provide list of external package dependencies and links to these packages
   • Masha will start working once release candidate is out
   • May have an intermediate release when distribution is ready
6. Hardware:
   • John built new csep-cert
     1. CruiseControl is installed
     2. John will install missing 3-rd party packages CSEP is using
   • csep-devel already uses new hardware, will need to update packages