Technical call

• Discuss performance task, including location sampling improvements and PSRC testing, representative logsums mini-spec
  • Dynamic chunking was an opportunistic improvement as part of TVPB work and could be even better
    • Probably could get 25% increase in performance with additional fine tuning
    • It experiments with a few choosers and then sets chunk size
    • The chunksize setting could evolve to be more of a "allowed_machine_memory_to_use" setting, which would be better for the user
    • It could better understand overhead memory usage
    • Probably need a couple weeks to implement the improvements
  • taz/maz location sampling coming together for PSRC
    • 28 hrs into the task
    • Need to implement for non-mandatory and trip models still
    • Found a few issues in PSRC data that PSRC is fixing
    • Also implemented skipping calculations if size term == 0, which will help all models, especially sparse ones like school location
    • Implemented in 2 and 3 model setups too
    • Maybe want to implement by mode since auto uses taz data but walk uses maz data
    • Need to double check on the shadow pricing math
  • representative logsums mini-spec makes sense
    • probably need a week or some more to implement
    • will really speed up the model and will help everyone
    • current activitysim model deployments by RSG can help fund if needed
    • for scheduling models, should be 10x speed up for MTC and 100x speed up for ARC
    • some lose in parking duration precision, but typically worth it, and feature will be optional
    • it better aligns demand and network model precision
    • suggest coding these settings in the network_los.yaml for one-stop-shopping of skim/time period related settings
    • also check when settings read in so end is after start and start and end are valid for the skim time periods
  • We decided to do just these topics - taz/map location sampling and representative logsums - with the task
  • And we'll create an issue to keep track of potential performance enhancements
• Discuss new activitysim.estimation.larch sub-package within ActivitySim
  • larch subpackage added to asim and potential for other estimation tool integration (like ALOGIT) also possible in the future
  • Tests not working since the latest numpy/pyarrow packages are inconsistent but the open source community will likely soon fix this so we'll wait
  • Jeff to keeping working on larch notebooks
  • I'll work on merging estimation, estimation_larch, and develop branches + merge Doyle's mp PR for populationsim
• Discuss adding test coverage for ALTS_PREPROCESSOR added for ARC next time
• Also let's review issues for any hot topics