Project Meeting 2018.06.08

Stop Models

• Update from @Jeff on trip purpose + destination model full model run results
• Models implemented and documentation about done
• Runtimes for the trip destination model are inline with the other (sampling + logsums for the sample) models
• Some trips are failing, which is most likely due to incomplete expressions in the upstream models. We've noted these and will return to them when it makes sense.

activitysim.core.pipeline - load_checkpoint table trips: 22,550,374 total trips

Time to execute trip_purpose : 103.635 seconds (1.7 minutes)
Time to execute trip_destination : 22703.613 seconds (378.4 minutes)
Time to execute trip_purpose_and_destination : 545.526 seconds (9.1 minutes)

### trip_destination 

activitysim.abm.models.trip_purpose - 9140321 last outbound trips
activitysim.abm.models.trip_purpose - 9140321 last inbound trips
activitysim.abm.models.trip_purpose - 4269732 intermediate trips

running trip_destination for 4,269,732 intermediate trips

model                           iter        failed
trip_destination                  0           1507    
trip_purpose_and_destination.i1   1            850    
trip_purpose_and_destination.i2   2            795    
trip_purpose_and_destination.i3   3            792    
trip_purpose_and_destination.i4   4            791    
trip_purpose_and_destination.i5   5            791    

• Update from @Jeff on trip timing model full model run results
• Models implemented and documentation about done
• Some trips are failing, which is most likely due to incomplete expressions in the upstream models. We've noted these and will return to them when it makes sense.
• Remember:
  • The trip timing model is using the input trip timing probabilities file used for TM1, which was created from sparse survey data and adjusted to work with the full sample of households when run through the existing TM1 implementation. This file has cells with 0 percent, which can lead to trips that cannot be scheduled.
  • We don't have all the upstream model expressions implemented, so the results below are for a somewhat funky set of trips. Since TM1 has been successfully run many times, we expect the remaining failed trips to largely go away once we finalize the upstream models. Plus, we added functionality beyond MT1 to re-run these models with new random numbers in order to resolve problem trips, and fixed the model to consider at-work subtour time constraints.

activitysim.core.pipeline - load_checkpoint table trips: 22,550,374 total trips

Time to execute trip_scheduling : 4158.853 seconds (69.3 minutes)

### trip_scheduling

trip_scheduling scheduling 22,549,583 trips

iter        failed
  1         362365
  2         207968
  3         140671
  4         106904
  5          87788
  6          76289
  7          68802
  8          63844
  9          60215
 10          57403
 20          47974
 30          45640
 40          44628
 50          43959
 60          43626
 70          43339
 80          43110
 90          42993
100          42787
200          42133
300          41945
400          41874
499          41769

• As we start to move from implementing modeling software infrastructure to implementing actual models with these components, the question arises about how we should resolve inconsistencies, and other design issues, in the architecture? The team generally agrees that a handful of fails can be dealt with using a simple patch/hack/etc, but if there are a large number of fails, then a more thoughtful solution needs to be implemented.
• Plan to start working on trip mode choice model next week
• Will use any remaining funds to go back through and clean up significant gaps in the expressions
• @Ben to draft list of significant expression gaps based on issues, progress report, and code review

Documentation / release

• Example model design and sub-model designs updated
• Sub-models hyperlinked flow chart integrated
• Updated listing of all sub-model input files
• Updated data schema and built auto-documentation tool based on our data pipelining technology
• Updated getting started steps
• Lots of little clean-up in the docstrings and existing documentation
• Plan to commit and release by end of day Friday
• Discussion postponed to next call

Project Management

• Updated Progress Report through trip scheduling
• PMC met to discuss scope and @Joe to revise amendment for the next phase of work
• We expect to complete the current scope and expend all funds by end of June
• Likely to execute amendment in September
• Planning ITM event for Monday evening