Agenda

• Admin Items
• ActivitySim Website Content
• Updated example naming/classification system #571
• Phase 7 task updates
• Memory Usage

Action Items / Decisions

• Reminder for partners to fill out the Model Inventory and Model Evaluation google sheets/doc. We may want to discuss results of this/next steps at a future meeting. Joe to decide.
• Each partner to send Caitlin content for the AMPO website (such as why you joined, examples/samples of implementation, any new logos, etc.), due August 11.
• Joe to confirm Jeff's proposed naming convention for models.
• WSP and RSG to code review Jeff's sharrow pull request, due August 11.
• WSP to produce a memo assessing ActivitySim's memory usage, due ?.

Meeting Notes

• Admin Items

  • Update on upcoming meetings, which can be found in the meeting notes
    • 8/2 – This meeting will be cancelled.
    • 8/4 – Wu to lead – update on escort model work from WSP
    • 8/9 - Rebekah to lead

• Update on contracts with AMPO. Caitlin will be provide a more comprehensive update next week. Everyone’s funding agreements are in except SF (which is coming) and Ohio. Update on written checks will come next week. Funding reminders will come out in the next couple days. Invoices have a 90 day requirement.

• ActivitySim Website Content

  • Follow-up to discussion on June 30 (https://github.com/ActivitySim/activitysim/wiki/Project-Meeting-2022.06.30)
  • Partners were going to update the AMPO website. Joe was going to ask for volunteers, but not sure if that happened or if anyone volunteered.
  • Caitlin shared the wireframe of the website. ActivitySim_WebsiteMockUp_JC.pptx
    • The new landing page will include a Why and a What and the Development Timeline.
    • The Consortium Page explains how the consortium is formed and how it runs.
    • The Github page links to github.
    • AMPO needs partners to update the Examples Page: why the agency chose ActivitySim, how they are using or will be using in the future, etc.
    • Each partner that has a “Production” or “Prototype” model (see below) should have their own Example Page, where people would be able to link to their example model. AMPO would like for each Partner to have a page, with at a minimum, why you joined and what you look to get out of it. Each partner should send Caitlin why you joined and what you want to get out of it by 8-11-22. Send Caitlin any updated logos. Caitlin will send the mock-up to Michelle.

• Visualization / SimWrapper update from SFCTA - Deferred due to Joe’s absence.

• Updated example naming/classification system #571

  • Want to formalize a system for classifying these examples, so that people know whether or not an example model is a real one in production, or close to being ready, or just an example/test that’s not intended to be used in real life (aka not calibrated/validated)
  • Jeff has proposed to have 3 classifications of models: Production, Prototype and Placeholder. See the github comment.
    • Production – these are real models. None currently exist.
      • Ohio won’t make their model public because it relies on QCEW, which is proprietary.
      • PSRC can publicly release their production model.
      • SEMCOG is close but they don’t want to make their data publicly available.
      • Note that one doesn’t need to public post the data but can post the code. They can put out the model but not include the data and replace with open source data that’s close, so it’s the still calibrated and validated
    • Prototype – not “official” but “ok” to use
      • Suggestion to rename to Example or something for the cases where it is the official model but not the official data.
    • Placeholder – computational testbeds, should not be used in application
  • This has general consensus, but will wait for Joe’s confirmation.

• Phase 7a Scope of Work

  • Task 1: Project Management ongoing
  • Task 2: Performance Enhancement Part 1 Sharrow for 1 zone system – All non-CS contractors are running out of budget and no one has done a code review on any of this. Sijia has looked at some of this, but Jeff has not received any actual comments in the pull-request. Documentation is still outstanding.
    • RSG wants to take a look at this but is unsure how much budget is left. Joel is on vacation, so the available budget isn’t available. WSP will start on the code review. CS would like code review done within the next 2 weeks. RSG and WSP will let us know on Thursday 8/4 how many hours they can devote and when they can have it done. Comments are expected to be submitted to github. All 1 zone models are now passing their tests.

• Task 3: Performance Enhancement Part 2 A pull request is open and all models on 2 and 3 zone systems are running and performance enhancements will be completed on 8/23.

• Task 4: Performance Enhancement Part 2 The Time Windows pull request will be open, submitted and ready for code review by 9/13.

• Task 5: School Drop-off and Pickup Model The pull request will be open, submitted and ready for code review by 8/15.

• Task 6: Disaggregate Accessibilities – Update 7/26 – Decided Option 2. (see Project Meeting 2022.07.26 · ActivitySim/activitysim Wiki · GitHub) The pull request will be open, submitted and ready for code review by 8/31.

• Task 7: Revised Shadow Pricing Mechanism The pull request will be open, submitted and ready for code review by 9/30.

• Task 8: Extend ActivitySim Code to Enable More Flexible Number of Tours and Stops This is a part of Task 5.

• Memory Concerns – See Project Meeting 2022.07.26 for last discussion

  • Options are:
    1. Joe proposed that we fix data types of skims immediately. The hope is that skims can be stored as integers to cut the footprint. This would require developing a new task order, issuing it and then doing the work. It would likely take weeks to get the task order and another set of weeks to do the work. The hope is that skims can be stored as integers to cut the footprint.
    2. WSP proposed to produce a memo that summarizes their assessment of the memory usage bottlenecks and then present the consortium with potential solutions.
  • To fix the data types immediately...
    • This is hard to do with the older models due to instability with the chunking, calculations and other data-type memory concerns.
    • CS does not have the staff availability to test this at this time without other Tasks falling further behind schedule. This is not a trivial amount of work. It is thought that Joe would like to see this sooner rather than later.
    • CS notes that the Sharrow implementation will solve many of these issues. However, model runtimes are not currently comparable due to problems with running the old models. The goal is that the legacy code can be stripped out of ActivitySim and these problems will no longer be an issue. There are still some issues that won’t be resolved for another year that require keeping the legacy skim formats.
  • In producing a memo assessing memory usage....
    • Wu agrees this is a good idea
    • Stefan agrees this is a good idea. Lots of preprocessing in the beginning that could be looked into, where data type is inferred but might need to be specified as integer instead of it assuming float. These are the types of things that need to be looked at.
    • All data types should be reviewed. There are also two more potential avenues that might be causing memory issue
      • Duplicative database carrying out through all the checkpoints in activitysim
      • ActivitySim doing vector maps when solving alternatives
      • Maybe the data types fix was proposed first because it would be the most straightforward to change, fewer code changes required. Knowing that it’s not the whole problem, but hopefully a good chunk of the problem.
  • It was decided that WSP will spend their 8 hours to produce a memo and do the Task 2 sharrow code review as discussed above.