Project Meeting 2020.11.19 - ActivitySim/activitysim GitHub Wiki

5b/6a Task 1 Project Management (milestone)

  • Cancelled call scheduled for Thanksgiving

5b Task 2 Strategic Planning (milestone)

  • Joe's revisions to V3 posted
  • Adding a bunch of comments that request: 1) a more complete articulation of the set of changes that people may wish to make; and 2) a more complete description of how the ActivitySim methods (exposing settings, etc) facilitate making these changes.
  • Reordering the sections to move "Managing contributions from others" above "Maintaining support for multiple model implementations"
  • Moving the test system discussion to the end of the "Maintaining support for multiple model implementations" section
  • Adding an example that starts to illustrate the types of changes that people would need to make to implement a significant enhancement
  • This is a living doc but we need to wrap up this version from a contractual point of view
  • So me, Joe, and others to setup a 2 call hour to go through the todos and get them done
  • @Joe setup doodle poll
  • Would be good to have new partners review as part of their onboarding/welcome

5b Task 10 Support for Three Zone Systems and TVPB (milestone)

  • Multiprocessing working with both initialize_tvpb (which builds the tap_tap_utility ‘skims’) and tour_mode_choice (and with all models for the little 3-zone test dataset)
  • Now working on tests and will commit later today/tomorrow and then I'll run the full scale example to check for correctness and performance
  • The implementation re-ranks/picks best paths over and over again like CT-RAMP
  • It does de-duplicate the ranking/picking within chunks though
  • Let's review the performance next week once we have something to react to
  • I plan to finish the comparisons to Marin TM2 as well once I have the updated version
  • For discussion, here's what the model steps looks like
models:
  - initialize_landuse
  - initialize_households
  - initialize_tours
  # --- single-process STATIC cache prebuild steps
  - initialize_los
  # multi-processable step to build STATIC cache
  - initialize_tvpb
  # ---
  - tour_mode_choice_simulate
  - write_data_dictionary
  - track_skim_usage
  - write_tables
  - write_summaries
 
multiprocess_steps:
  - name: mp_initialize
    begin: initialize_landuse
  - name: mp_tvpb
    begin: initialize_tvpb
    num_processes: 4
    chunk_size: 0
    slice:
      tables:
        - attribute_combinations # time-of-day, demographic segment, access mode combinations
  - name: mp_mode_choice
    begin: tour_mode_choice_simulate
    #num_processes: 2
    #chunk_size: 0
    slice:
      tables:
        - households
        - persons
        - tours
  - name: mp_summarize
    begin: write_data_dictionary

6a Task 2 Complete Estimation Mode (milestone)

  • Newman working on the run_all_steps feature with the revised one EDB for all person types for non-mandatory tour frequency interaction_simulate

6a Task 3 Telecommute Model in Cooperation with SEMCOG (milestone)

  • Telecommute design and links to background docs posted
  • Drafted example shared Tuesday, which had two new models - work_away_from_home and telecommute_frequency
  • We'll use the name work_at_home instead of work_away_from_home (which is the other alt in the binary logit model)
  • Currently thinking about how/where to do the work_at_home model in the model steps. There are three ideas currently:
    • before work location with an aggregate accessibility measure for employment access
    • within work location with an extension model so we can get the mode choice logsum for employment access
    • after work location since the mode choice logsum will have been calculated, saved, and therefore available
  • SANDAG has a binary logit model before work location
  • DaySim has a couple ways of doing it - a separate model and a nested work location model with a work_at_home nest
  • PSRC ran into shadow pricing issues with the nested option
  • Separating it out may provide more flexibility/clarity to users
  • Should we use a simpler aggregate employment accessibility term or multimodal logsums? It's both a model design and plumbing question
  • The bay area TNC survey suggests Uber drivers have a more complex day pattern than a combined work_at_home/location model so maybe separating this model from the work location model provides more flexibility for future revisions
  • If run work_at_home after work location, then set the work location to 0 for workers who work_at_home
  • I'll talk with Mark and Joel and they can join the next call to talk about this submodel, but also the overall project goals/challenges with adding new submodels/example data/etc
  • Its important that we be clear about exactly what we mean by telecommute, work_at_home, etc.
  • Maybe just implement the SANDAG version for now and make it better later?
  • Once we get this working, do we extend the existing tm1 example with dummy data? Do we want to maintain that data into the future? This is basically the same issue we discussed with Clint related to new models being built and tested with ARC data.

6a Task 8 Maintenance and Support (milestone)