Project Meeting 2022.06.09

Admin

• Vehicle Type Models review is complete and merged to develop. All Phase 6 work is complete.
• PR #571 changes name of MTC model to Gondor, to make clear it's not the MTC model and not any kind of "official" model of any real place.
• Ready to package up and release version 1.1.0

CDAP and Joint Tour Frequency

What are the changes?

• Joel presented on the reason for changes and overview of proposed changes. CDAP and Joint Tour Frequency Models.pptx
• Background – joint tours tend to be later in the day, compared to observed data. Because the mandatory tour scheduling, they don’t know anything about joint tours. Not enough overlapping time windows in the household for joint tours to occur.
• Existing Model
  • Assessed at the household level, up to 5 persons, identify M, N, or H for each person in the household.
  • Person type interaction terms
• Updated Model
  • Extend CDAP to consider whether a joint tour is generated at the household level.
  • If at least two people in the household are making a tour, then there is a choice for joint tours.
  • Allows for workers to leave work early, to make time for joint tours later in the day, for example.
  • Is there any reason someone would not want to implement this: data requirements, run-times, have you demonstrated that the new structure allows for better calibration/scheduling, etc.?
    • Fewer joint tours later in the day has been observed with new structure
    • Need to transfer coefficients to time of day models
    • Run-time implications aren’t significant because these models don’t take long to begin with
    • Only reason to not implement is that it makes the model more complicated, harder to explain

How are they implemented?

• Sijia presented on the implementation of those changes. MTC ActivitySim Implementation Code Changes.pptx

• Additional term added to the utility function for joint tours, with household variables and person-type/day-pattern interaction terms.

• Model is still applied at the household level and indicates if there is a joint tour or not

• Change 1: Added Joint Tour Utility Extension switch in cdap.yaml, to turn on or off – default is false. If you turn to true, then you need to make sure you have the right coefficients, that include join tours.

• Change 2: change the alternatives to include alternatives with joint tours

  • Coefficients are the same as the non-joint utilities
  • But then add the joint tour utilities, coefficients are only for the alternatives with joint tours

• Change 3: Calculate joint tour utility in cdap.py, adds in the joint tour utility

• Change 4: New alternatives for joint tour frequency and composition model

• Individual non-mandatory tour frequency

  • No structural changes but caught a few bugs
  • Hard-coded model specs,
    • Maximum number of tours – will crash if the number of alternatives is more than hard-coded max
    • Type limitation to int8 for alternatives

• What does the user need to do:

  • Set the Joint tour frequency composition to run in settings.yaml, comment out joint tour frequency and composition

• Discussion about whether developing a switch to turn on/off some enhancements is a good idea

  • Add complexity
  • Potential downstream impacts may be challenging, especially with very complicated with things like the vehicle type model, but less complex things like this particular change may be doable

• What would be the next step now that the consortium have heard about the CDAP model changes?

  • Not enough time to discuss, so will be the topic for next Tuesday’s technical meeting: What to do when new stuff is being introduced/developed?