Project Meeting 2021.02.23 - ActivitySim/activitysim GitHub Wiki

Technical Call

Plan to get started on representative logsums soon
Review issues
- Jeff to fix #386 and look into #380
Update from Doyle on tidying up agency examples
- Now have PSRC, ARC, SEMCOG, and Marin in testable form:
  - Cropped so portable/runnable
  - Cleaned-up any issues with running with the latest code (and documented in a change log)
  - Have a test to test for successful model run
  - Travis now runs mini/cropped version
- Built "cropper" tool
- PSRC example exercises chunked accessibility code (but not for TVPB since PSRC is a two zone model)
- Jeff to push code to develop2 and then share code with Stefan and Wu to test
Update from me on the task 6 prototype multiple models test system
- I'll keep working on it based on feedback and we'll discuss again next week
- Part of the cost of AWS is automation to avoid excess costs
- AWS great for common access
- Need to use consistent language and terms
- Could put on an AMPO server as well
- Could amortize cost over annual budgets
- Need to better define use of future warnings
- AWS has an instance scheduler that PSRC is investing
- We run mini/cropped version often and full version less often but at least with agency update
- How much does the consortium want to be in the model correctness business?
- There's multiple levels to this - does it run, is it perfectly correct, but also how's performance/scaling, and other in-between states
- What's the use case to get new features into the core?
- Let's discuss/add the ARC example to help explore the ideas
- An additional goal is to make it easier for additional agencies to more easily adopt new methods
Update from Newman on estimation work
- See progress on issue #382
Clint will share the ARC implementation next week