Agenda

• Phase 8 wrap-up check-in
  • CS PR for Component Documentation (check-in on WSP review)
  • WSP PR for Data Type Optimization
  • RSG PR for Input Checker
• Phase 9 Scopes of Work and Contracting

Meeting Notes

• Phase 8 wrap-up check-in

  • CS PR for Component Documentation - WSP is still reviewing PR and coordinating with CS on comments.
  • WSP PR for Data Type Optimization - Jeff is still reviewing PR and coordinating with WSP on comments.
  • RSG PR for Input Checker - Jeff is still reviewing PR and coordinating with RSG on comments.

• Phase 9a optimization task process

  • Initial issues
    • Issue 1: sharrow performance in workplace location choice (district-specific constants)
    • Issue 2: run time for vehicle choice model
    • Issue 3: school escorting taking a lot of memory

• Communication/process

  • Propose to use GitHub for tracking collaboration/management of task
  • Propose to create repositories for the protype models (not where they are now, in the examples), which will break apart source code and implementations.
    • Agree to divorce the SANDAG repo with the source code but know that they will have to come back together.
    • Sometime in the next two-three weeks, we need to stand up the SANDAG prototype model.
    • If we want a real before and after comparison, we can’t change anything in the configs, params, etc.

• How do we memorialize that the two examples that we do the optimization on? These are the official models that are going to be the ones that we maintain. However, the SANDAG model is still under development.

  • For this scope of work, we need to draw a line in the sand and keep the models static.

• There are extensions that are not part of the core ActivitySim code (such as the transponder ownership and external/internal workers models). How do those fit into this? How do we treat bespoke models?

  • Roadmap addresses some of this, by describing the two categories of simple models. A few lines in a config could create any logit model.
  • Agreement that there’s no issue with including the extensions in the SANDAG model optimization.

• SANDAG model can only run on 1TB RAM machine, requiring high resources to run/test it. They haven’t tried chunking yet. However, the starting point will include the data type optimization, which should reduce memory requirements.

  • If the initial run results are – “I can’t even run it,” should we have a scaled down model version? Joe doesn’t want to do this. SANDAG has been able to run the entire model with a 1 TB RAM machine, and it takes about 4 hours.

• Preference to use the term “BayDAG” instead of SANDAG for the canonical example.

• Will take 2-3 weeks to get started because of the required following steps:

  • Need to wrap up Phase 8 PRs
  • Need to approval all the BayDAG PRs
  • Then need to merge all

• What can be done now

  • Discuss issues, things they plan to do, etc.

• Engineering team will meet at the regularly scheduled ActivitySim Tuesday meeting time, but consortium members need not attend. If they would like to listen in, they are welcome to or watch the recordings. For next Tuesday's meeting, the engineering team will:

  • Go through list of all the PRs, work towards getting them merged
  • Start using the GitHub issues as task lists and assign them to people, add people into the discussion – use it the way it is intended to be used.
  • Go through backlog of issues, find the ones that are serious things, especially those related to performance. Create new issues if something’s missing.
  • There are labels that can help the group prioritize which ones to go through.
  • Develop level of effort estimates for issue fixes (related to performance) to help inform prioritization.
  • Use the project board to pull in the issues that they are going to work on.