Agenda

• Code Review Updates
• Memory Profiling Updates

Meeting Notes

Code Review Updates

• School escorting and flexible number of tour and trip IDS
  • RSG is working on school escorting comments from Joe
  • RSG has addressed flexible number of tour and trip IDs comments from Joe
• Estimation and Random Seeds
  • RSG has estimation fix that they are working on
  • RSG submitted comments to Joe on random seeds
• Disaggregate Accessibilities
  • WSO provided comments to RSG
• Sharrow
  • RSG comments are in, and they are done from their perspective
  • WSP is currently working on their review

Memory Profiling

• Follow-up on action items from last week
  • Data types
    • WSP found a place in the settings yaml where users can specify the type of data. Tested changes with 25% sample ARC run but there was not much memory savings (approximately 26 GB memory reduction), less than saving that originally expected. Only updating input data type doesn’t help that much.
    • It is suspected that When data is used in combination, the model defaults to the higher precision, so that may be what’s going on, but unproven/tested at this point.
    • ACTION ITEM: WSP to update the data type usage table, adding in the data rows for each of the files to get a sense of data size when you look at number of rows and precision.
    • Focus is on the data types because it seems like this may be easier to fix. However, it is unclear if there even ways to fix this efficiently? Joe remembered something that Jeff said about how difficult this might be. ACTION ITEM to check with Jeff on this.
  • ACTION ITEM: WSP to upload tableau workbook to tableau public.
  • Next steps
    • Mandatory tour scheduling is biggest memory user and WSP can look into ways to reduce memory usage of this component. Current options include:
      • Reducing the number of alternatives for those that are highly unlikely (such as work tours starting at 3am). WSP to look at what RSG did for SECOMG in their scheduling component.
    • MWCOG example is currently crashing for WSP. MWCOG will help Sijia to get a working version of their model. They have a separate development repo that works for them that they may suggest using, instead of the ActivitySim release.

Other

• Chunking
  • This was something implemented to solve a problem, but was never an ideal solution.
  • Chunking addresses some initial design issues and some asked if we need to go back to the original design to see if there is a better implementation that wouldn’t require chunking.
  • Joe wants to look for lower hanging fruit first before considering a big refactoring process, for example.
• KPMG did a security scan on the ActivitySim code (for a Sydney project) and identified some issues, some of which are easy to fix and others require more work. report is shared here:DetailedReport_ActivitySim_1_Nov_2022.pdf