Project Meeting 2020.05.15 - ActivitySim/activitysim GitHub Wiki

Task 1 Project Management (milestone)

  • Discussed Phase 6 scoping and ranking here
    • Team discussed updates to task descriptions included under feature descriptions + revised descriptions
    • Collaborating with SEMCOG on the telecommute and transit pass models makes sense
    • Tasks that go into significantly new modeling territory should probably start with a review/design task and be phased since they likely consume too much resource in each phase
    • Maybe ARC time-of-day choice model is an incremental step toward time-of-day improvement without the large scale revisions required for the DaySim approach?
    • Adding a household vehicle table is a foundational first step toward more involved vehicle modeling
    • Partial automation of calibration could mean writing a calibration data bundle
    • Airport model maybe use SANDAG as the example?
    • Add more detail to the transit pass plan
    • Better articulate the testing space
    • Important for partners to download future updates to the software and it just generally works
    • Support for partners yes, support for others to a lesser extent
    • Next steps - @Ben add some more detail and add rough LOE estimates
    • We discuss next time (5/22) and then PMC prioritize by next call (5/29)
    • Then draft scope of work first week of June
  • Let's leave PRs open for a week or so so everyone has a chance to see/review
  • We need regular releases and release notes
  • FYI, SEMCOG full sample single process run works but we're investigating a multiprocessing issue

Task 9 Estimation Mode (milestone)

  • Cleaning-up implementation, including some small issues in expressions revisions, improvements to settings for saving logsums, expression tracing, etc.
  • Starting user documentation, see new flow chart of process that I shared today
  • Leaner/cleaner data model improvements happening as well
    • Can now identify input fields to keep on load and many in the example are not needed so the table sizes are smaller, which means less disk space for the pipeline and the ability to fit more households in RAM, which should help with runtime
    • Pipeline steps with underscore at the beginning of their name can be run without writing to the disk, which means using less disk space and a bit faster run
    • Still need ability to save select fields such as costs for later analysis
  • FYI, setting up functionality for SEMCOG so this is helping identify improvements