Project Meeting 2022.10.06

Agenda

Consortium to review [issue 609](o https://github.com/ActivitySim/activitysim/issues/609) on in-commuting and respond to that thread to reach a resolution by next Thursday.
Consortium to email questions/comments to Ali ([email protected]) regarding about the mandatory location choice model and Nick ([email protected]) about the disaggregate accessibilities task.

Tests work on Mac and Windows but Github Actions runs on Linux. It doesn’t pass Github Actions because it how multiprocessing works in Linux, RSG is looking into this.
Implementation
- Note that the target threshold is different from size threshold.
Results
- Previous showed SANDAG 2-zone
- Now showing MTC one zone
- Run time – simulation method takes about 20 minutes
- Differences in average distance by iteration is greater in one-zone system because reallocation of simulated workers are moving around TAZs, which are bigger and further apart, versus SANDAG’s 2-zone system where the zones are smaller.
- Increase in half-tour length has implications in calibration, need to include simulation when calibrating.
Methods to adjust input employment data to factor in in-commuting
- https://github.com/ActivitySim/activitysim/issues/609
- Instead of discussing during this meeting, please see the issue referenced above and response within the issue by next Thursday. If there is no resolution, RSG will implement something.

Functioning model working for 1-zone and 2-zone
Working on multiprocessing tests but ran into some issues
Process is currently broken into two steps: initialize the proto population and then to compute disaggregate accessibilities
Merging step is a recent recently completed task, involves a one-to-many join where user specifies the columns/fields to join
Potential issue of unique household types in full synthetic population that doesn’t exist in the prototype. Nick explored joining something close but requires more user-defined specs. This is something that could be discussed more.