Project Meeting 2022.05.19 - ActivitySim/activitysim GitHub Wiki
Meeting Notes
Admin
Invoicing to partners has been delayed, but expected soon.
Metropolitan Council Update
Last year, MC set up a basic model using ActivitySim, starting from the SEMCOG model as a donor model.
No calibration or estimation, but some basic adjustments to align with the recent TBI survey data.
MC wants to talk with other agencies about what to do next, given their timeframe and budget for the
next phase. Ideally by June.
Performance: the model takes a day and a half, two days to run.
Econometrically consistent simulation of multinomial and nested logit choices
A proposal by @janzill with VLC in Australia. He opened this new issue #568
to discuss.
A new method to make simulated choices would be created: instead of making one random draw from a spectrum of probability, we would
make one random draw per alternative.
In the limit for infinitely large populations, this change in theoretically meaningless, but all actual simulations have a finite population, and the change could result in practical changes.
This feature would be activated by a settings / yaml configuration switch.
It is expected to have little runtime implications, neither notably faster or slower than the existing system.
The consortium generally approved of moving forward with this idea.
Some concerns were raised about whether and how this might impact the generation of logsums for upstream components.
Would we (could we, should we?) pass the drawn maximum utility up to higher level models, instead of logsums? Probably not for now, but this is worth thinking about.
Everyone is invited to post comments/questions into the GitHub issue.