Skip to content

Project Meeting 2019.05.31

Ben Stabler edited this page May 31, 2019 · 18 revisions

Verification

  • See the verification table for the current summaries
  • Updates summaries for shadow pricing since we also ran a 100% multiprocessing example
  • Issues we fixed since our last call, which are in pull request PR#261
    • Updates to non man tour freq expressions
    • Updates to atwork tour freq expressions
    • Updates to tour stop freq expressions
    • Updates to tour and trip mode expressions
  • The verification results of these submodels are better as a result
  • Added 100% sample multiprocessing example run setup to the user documentation
  • Published (released) an updated version - https://pypi.org/project/activitysim

Planning Applications preparation

  • Parallelization presentation
    • Refreshed the full scale multiprocessing runs for our presentation
      • 28 core Windows server 165 minutes
      • 64 core Windows cloud server 105 minutes, $12
      • 128 core Windows cloud server 84 minutes, $27
      • 64 core Linux cloud server 71 minutes, $8
      • 128 core Linux cloud server 51 minutes, $16
    • OSX and Linux running faster than Windows
    • Python 2 running faster than Python 3 due to multiprocessing shared data issue that we're investigating
    • Need to experiment more with Intel MKL library settings on Windows
    • Updated the Azure cloud run setup scripts
  • Workshop
    • @Joe working on beginning and end of workshop slides
    • @Ben on platform and runtimes
    • @Bhargava on setup, running the model, sensitivity tests
    • @Brice on outputs, including Tableau dashboards
    • @Joe/Brice add some verification slides
    • @Bill to close session and welcome follow-up
Clone this wiki locally