Project Meeting 2020.12.08

Technical Call

• Strategic plan
  • Meeting time poll sent to partners last week
  • Plan to get feedback next week
• Update from Doyle and Newman on estimation improvements
  • non_mandatory_tour_frequencey EDB reformatting is done
  • the CDAP model has a different configs file organization and we're going to match it up better with the others, like tour mode choice
  • Basically replace the coefficient numbers with named strings for larch integration
  • larch not reading the processed files - it is reading the input files since it needs to create the input files too after re-estimation
  • Should wrap these up soon and then on to the trip model improvements
• Update from Doyle and me on verification of TVPB
  • Tidied up some loose ends in the chunking calculations
  • Cleaned up some documentation
  • Now that we have dynamic chunking, maybe we have users specify a RAM chunk size in GB since its more user friendly?
  • SEMCOG project is going to test the new chunking capabilities
  • I'll run a full scale TM1 example too
  • PSRC to test and the code is on the multizone branch
  • Set a chunksize and then ActivitySim will override with a dynamically calculated value
• Work after estimation improvements
  • Doyle should probably work on performance
  • Could be the logsum O,D,duration de-dup / representative duration task
  • The idea is to de-dup not just choosers but also alts
  • Could also be string / factors
  • Probably have Doyle investigate a few items and pick the most promising
• Update on TVPB verification
  • I need to get back to it. Plan to share some results next week.
  • Make sure no logic problems
  • Summarize some select trace results and also some aggregate results
  • Will share tap-to-tap utilities trace, maz-to-maz results, tour mode choice, and regional tap counts summaries
• Continue discussion of Berkeley/LBL template model and simpler example ideas
  • Maybe tm1 is not too complicated, maybe its just not the right starting point for their purposes
  • It seems a simpler ABM example model would be really helpful here, and for other academic projects
  • Templating model designs sounds cool, but it's probably too much for this project at this point
  • We want a model that is super tight and fast
  • We need to be strategic with our resources and exploratory work isn't core to the mission currently
  • A simpler example is probably super useful for the growing community
  • Divergence (fracturing) in model design/software has been our industry's issue so let's keep it focused
  • Can a university create the simpler ABM example for contribution?
  • This is a good opportunity for a pull request
  • Template a bad idea - its trying to make a very specific software generic; its better to make a simpler version
  • Lots of interest / support for an activitysim lite version
  • It needs an owner / maintainer, maybe Oregon DOT?
  • Promising owners are Oregon DOT, Ohio DOT, Caliper/Central Coast, a University
  • Consortium will help and should maybe add to our scoping effort next time around
  • There are partnership costs to setup a new spine and maintain
  • Will talk with Greg Macfarlane at BYU next week during this time