Joe working on revisions and plans to share later today
5b Task 10 Support for Three Zone Systems and TVPB (milestone)
I'm now running the Marin TM2 full scale example and reviewing results for correctness
I'm using the new tour file loading from disk functionality and it works well. It is consistent with the existing functionality (restarting and estimation)
Would be good to add a trip loading from disk function too (maybe under completing estimation mode)
The big outlier tap is Richmond BART station so its related to no PNR trips as well
The Richmond BART station in Marin TM2 has 6200 PNR trips and 0 in asim currently
Next step is to trace a HH that selected PNR via Richmond BART and review and fix calculations
4 summaries for validation:
aggregate mode share
tap counts
tap-to-tap total utilities
for some HHs, paths considered, paths rank, and best N paths finally considered as tables/maps/something
How much info/tracing can we get from TM2? Asim has a lot
I'm running the model single threaded
It took 3 hours for 775k work tours
Two zone system setup QA/QC
We should do more to QA/QC the two zone setup code
I could share the setup on a later call? We could trace some blending calcs, some results from the toy example, etc.
What about DaySim park and ride lot choice? Maybe functionality added later
Maybe we wait for PSRC to get further along and then use a real example for QA/QC?
PSRC is considering different number-of-zone-system configs for different users since many users don't need the additional complexity
Let's add documentation to the user guide on the zone config options and pros/cons:
Goodness of results
Sensitivity
Flexibility
Runtime
Data / network management
Consideration of assignment software capabilities
It would be super cool to get a research project to compare the three approaches using activitysim
Jeff working on multiprocessing performance and caching strategies
The existing shared memory data objects for multiprocessing are tricky
It is more low level programming than traditional python and so improved methods for sharing data across processes, like maybe caching, would be a good improvement for maintenance and developer burden
We can discuss in more detail next week if we want
Tuesday presentation and discussion was good, especially since we're moving from re-factoring existing functionality to creating new more undefined functionality
Lots of discussion around replacement activity and AWD models versus Monday/Friday telecommuting patterns
We don't have explicit telecommute program sensitivities, just the ability to change constants. That's ok, but more difficult to explain
SFCTA, PSRC, SANDAG and others will discuss existing experiences with telecommuting analysis with DaySim and CT-RAMP
SANDAG did VMT analysis under various telecommuting scenarios