Project Meeting 2019.03.08 - ActivitySim/activitysim GitHub Wiki

Verification

  • We fixed all the verification table issues and re-traced a HH to confirm the fixes
  • TM1 style shadow pricing looking better, see plots below, but still not converging as fast as TM1 so we're still investigating
  • Size term by iteration for zone 1 work low income segment below
zone 1, work_low, sp iteration ctramp style
target 3657
1 2490
2 2625
3 2824
4 2860
5 3069
6 3132
7 3123
8 3188
9 3249
10 3346
  • Daysim style shadow pricing needs some fixing as well (after reviewing the code and discussing with Mark), which we are still working on, see #250. We also updated the expression file as a result.
  • Updated the example to use ctramp method and more efficient damping factor
  • Implemented mode choice logsums for mandatory tour time-of-day choice as part of the verification/clean-up
  • Now working on a 10% sample run of the subsequent submodels, but ran into tour mode choice sampling and logsums not being implemented for escort tours.
    • We agreed to go ahead and implement missing agreed upon features such as this
    • This will slow down the model so its important we complete it before talking about runtimes
  • We plan to continue to trace calculations expression-by-expression to ctramp and to add expressions to force choices for the trace household to match TM1 so we can compare utilities
  • @Ben to create shadow pricing total workers by taz target versus modeled plot like ARC
  • @Ben to continue debugging shadow pricing calculations
  • @Jeff to implement logsums and sampling for non-mandatory tour destination choice
  • Fixing the daysim approach to shadow pricing isn't important so we'll either fix it soon or drop it

Project admin

  • We've spent about $145k of the $210k budget to date, ~$65k remaining
  • Our progress report is really starting to look complete
  • As we discussed at TRB, we prioritized verification over support for two zone system and assistance to ARC/SANDAG. These tasks were budgeted at $80k total so we're now into those budgets.
  • We decided:
    • @Ben/Jeff to estimate cost to implement remaining agreed upon features in our progress report
    • We will push the two zone system task to the next phase
    • We will push the ARC/SANDAG support tasks to the next phase
    • We will implement the TM1 free parking parking
    • Remaining budget will be used for implementing missing features and verification since our summer release is top priority
  • The updated DBE GFE looked good to AMPO

Shadow pricing by iteration scatter plots

  • c=ct-ramp multiplicative method, d=daysim additive method
  • gradeschool_c_sch_.gif
  • gradeschool_d_sch_.gif
  • highschool_c_sch_.gif
  • highschool_d_sch_.gif
  • university_c_sch_.gif
  • university_d_sch_.gif
  • work_low_c_wrk_.gif
  • work_low_d_wrk_.gif
  • work_med_c_wrk_.gif
  • work_med_d_wrk_.gif
  • work_high_c_wrk_.gif
  • work_high_d_wrk_.gif
  • work_veryhigh_c_wrk_.gif
  • work_veryhigh_d_wrk_.gif