Project Meeting 2023.10.10 - ActivitySim/activitysim GitHub Wiki

Agenda

  • Admin
    • Metroanalytics meeting (agency experience)
  • Data Type Update
  • Input Checker

Meeting Notes

Admin

  • Another Phase 9 scoping meeting for the partners this Thursday
  • Getting ready to write up the IRFP, likely to be released early November
  • Joe met with Metroanalytics, who is asking for agency experience. Joe will be sending an email to partners, soliciting volunteers to respond to this request.
  • Update on the response to Eric Miller editorial
    • Still waiting to hear back
    • Joe to check on editing privileges if anyone wants to add their names

Data Type Update

Presentation: ActivitySim Data Types Task - Progress Update 10-10-2023.pptx

  • Sijia compared two runs with and without tracing. Without tracing, the large spike goes away.
  • Previous reports comparing run time and peak memory all had tracing turned on. Sijia created another baseline run without tracing for a better comparison of the impacts of the data type update.
  • Additional test was run with Jeff’s changes to clear flow cache. However, it’s not dropping memory as soon as expected, and Jeff and Sijia are looking into it.
  • Sijia thinks she’s done all she can do related to data types. There may be more changes to related to sharrow to release the cache and potentially other non-data type issues.
  • Moving forward with what's implemented – the move from strings to categoricals is within the code, and the user doesn’t really need to define anything up front to make sure strings are converted. Everything happens in the source code. Longer term, the data model should make things more clear.
  • Next steps:
    • Code clean up and making sure the CI tests pass
    • Documentation
    • Sijia and Joe to connect separately to discuss how to wrap up this task
  • Questions about how tracing can be improved
    • What do we want tracing to report?
    • It would be doable but would require a lot of work. You would need to tell numba what to hold, instead of holding everything.
    • TO DO: Jeff to create an issue related to this task.

Input Checker

  • SEMCOG took the example input checker and guidance and then adapted for their model.
  • When SEMCOG ran it, they got a lot of errors in the log that weren't very helpful.
  • RSG is working to reformat the error output so it’s easier to read.
  • Next steps
    • Circle back with SEMCOG after changes have been implemented.
    • See if they can pull the SEMCOG checks into the canonical example.
    • Finish documentation
    • Check-in again in 2 weeks