Skip to content

Project Meeting 2022.09.15

mnbina edited this page Sep 15, 2022 · 3 revisions

Agenda

  • Next steps on trip scheduling issues

Meeting Notes

Admin Items

  • Phase 8 interim scope/budget for Phase 7 code review
    • Before, technical tasks were deliverable-based and PM/code review billed as T&M. Propose to keep the same but can discuss further at future Phase 8 scoping meetings between the partners.
    • Proposed schedule for interim task orders would last through June 30th.
    • Propose budget assumes 4 hours of software code review per week, about 160 hours or up to about $35k for each consultant, which is more than previous contracts. Joe senses that we’ll be more focused on housekeeping and core activities moving forward, instead of enhancements, which would require more on-call activities from consultants. Joe to discuss this at next partners-only meeting.
    • Consultant-specific add-ons would include:
      • Memory profiling for WSP would be included in WSP’s interim Phase 8 TO.
      • A limited number of hours for CS to continue project mgmt for the few months until Phase 8 scoping/bidding/awarding phase is complete.
    • Budget does not include attendance to all meetings. This would assume 4 hours per week x 50 weeks x 2 representatives from each consulting firm, that’s about $130k for just meeting attendance. If people have technical tasks, meeting attendance should be included in their deliverables-based budget to present findings.
      • Suggestions to reduce meeting budget spent by consultants:
        • Specify code reviewer at the beginning of technical presentations so that not all consultants need to attend and only the reviewer needs to attend meetings to understand the changes.
        • Fewer meetings – to be discussed at partners only meeting.
  • Sharrow’s code review by WSP is complete
  • Jeff pointed out bug that effects estimation and calibration, RSG to take a look.

Shadow Pricing (revised shadow pricing test results)

  • Presentation: RSG_revised_shadow_pricing_tests&results.pptx
  • Ran tests on SEMCOG model
  • Convergence criteria borrowed from CT-RAMP/DaySim
    • Thought DaySim could use absolute value or %, to account for zones with very small employment
    • Forgot to add in the presentation that small zones (less than 10 jobs) are excluded from the convergence criteria calculations
    • Configurable in the shadow pricing yaml
    • Max number of iterations is currently 10, but can be specified by the user.
    • Process automatically iterates until reaches convergence.
  • Previous flatlining convergence at 3 iterations is no longer a problem. The issue before was that sampling was occurring in the multiprocessing so it couldn’t converge. Now the sampling happening in the aggregate rather than the multiprocessing. Needed to add a new shared data structure for the threads so that the worker choices are available across all the threads after they each completed.
  • Run time was approximately 50 minutes, for the 9 iterations it took to reach convergence, which does not include reading skims and other processes not related to the simulation.
    • This is about the time it takes in DaySim; however, once it’s run, then the shadow prices can be kept for future runs.
    • Jeff noted that this process could/should run a lot faster with sharrow.
  • Next steps
    • Separate pull requests based on when the work from home module is run, whether in-commuting and/or out-commuting is implemented
Clone this wiki locally