Project Meeting 2022.08.25

Agenda

• Admin items
  • New docs includes versioned history, feature branch capability
  • End testing with Travis, migration to GitHub Actions is ready
  • Need for additional data validation checkers
• School escorting code review - postponed topic
• Shadow pricing update

Meeting Notes

Shadow pricing update (approach and implementation)

• Presentation: RSG_revised_shadow_pricing.pptx
• In destination choice, shadow pricing is a constant to doubly constrain the model. This is a simulation-based approach.
• Scaling term in order to account for the fact that workers does not equal jobs in the model inputs
• Added shadow pricing settings in a new yaml
  • Set method: “method3” but name could be changed
  • Set the land use file fields for scaling
• Shadow_pricing.py includes the iterating and most of the shadow pricing code
• Location_choice.py includes the sampling process
• Status
  • Code is mostly in place
  • Tested with a small sample from a 2-zone system
  • Currently running without errors
  • Number of workers to simulate plateau at iteration 4, 5, and 6 due to multiprocessing – currently testing without multiprocessing
• Questions
  • Since shadow prices cannot be estimated, and calculating new shadow prices can be time-consuming in terms of computational time, should shadow prices be updated when the modeled transportation improvements are significant enough to modify the usual workplace distribution?
    • Judgment call
    • Purpose of this process to make it faster so it’s easier to rerun if needed
  • Tests after implementation: how do things change iteration to iteration and in the end, look for potential bias. Metrics could include:
    • Compare flow from old method, new method, and observed
    • TLFDs, average trip lengths
    • Segment by zone size
    • There is documentation about this process but maybe not enough detail on the tests, could be expanded if requested (RSG edit: documentation is not in scope but there is a write up in the tasks that can be expanded upon to add detail on testing that will be conducted, or we can just continue to use slide decks as we go)