Project Meeting 2023.08.29

Agenda

Provided a summary of estimated memory savings if we downcasted integers
- Ran tests to make an informed decision to determine where to best spend resources – is it even worth trying to downcast numerics
- Numeric downcasting is complicated to implement and more likely to cause problems; without demonstrating significant benefits.
Optimization is working but not getting quite as far as we had hoped
Next steps:
- Continue to look into sharrow implementation (re time periods in categoricals)
- Explore “safe” downcast of numerics and see how much budget is left after that
- Dig into mystery spikes, to see if you can understand what’s going on
- Another check-in in two weeks

Presentation: input_checker_pt5_progress_update.pptx
Recap of previous presentations and work
- Previously presented on pandera vs pydantic and went with the pandera approach
- Each implementation would need to specify and write the checks that they want
Under the hood, it’s just these three lines of code
Data model support includes specific enumerations of variables in enums.py
All of this is included in a data_model folder
Setup and Implementation
- Include the input checker in the list of models, at the beginning
- Specify the name of the data model folder
- Check log file for errors and warnings
- There is a yaml file that specifies the name of the python file in the data model folder and list of tables that you are checking
Next Steps
- More testing and clean up of code
- Produce a list of checks included
- All of the checks in Oregon and SANDAG are in there now (except the network checks)
- Integrate into ActivitySim’s continuous integration test system
- Documentation
- Check back in 3 weeks

Partners only meetings are still discussing various things internally – roadmap and phase 9 procurement
Joe has sent a series of three more partner only meetings for the following three Thursdays