Phase 6a Update

• Chunking
  • I ran a few more chunking improvement scenarios for discussion
  • Its super important we keep on top of runtimes since reducing runtimes is a key part of the project
  • Chunking batch processes chunks of choosers when there's not enough RAM to solve the entire problem at once across all processes
  • Other components constrain the runtimes as well, like the cache
  • CPU processing speed mattered a lot for ODOT so ODOT to try their machine and report back
  • Let's try a couple more runs and then add the share of time for the HH submodels since this is the key part of the multiprocessed runtime
  • It's not part of the current work effort, but it would be great to do a more systematic analysis to inform agency needs about how much hardware to buy
  • I'm working on updating the documentation and examples for release
  • Matrix load times can be slow for big models so a multithreaded loader would be helpful
  • It runs faster on Linux so one can use a virtual machine (like docker) on Windows as well
  • Jeff to soon work on the additional two usability features
• Work from home
  • SANDAG shared the aggregate accessibility model results for comparison
  • Need to discuss with SEMCOG team how best to update/approximate for the SEMCOG example
• Transit pass
  • Joel shared SEMCOG transit fare discounts analysis
  • Will code into SEMCOG example next. Person type and operator skimmed fare discount percents will be added as user defined constants in the yaml files and then referenced in the expression files
  • Interesting to note that almost all subsidy was either 100% (free) or not at all (full fare)
  • Will share scripts if interested in doing this in another region + MTC is talking with Joel about doing the work in the Bay area as well
• Release
  • Working on tidying these items up for release as version 1.0