Project Meeting 2017.08.18 - ActivitySim/activitysim GitHub Wiki
Logsums
Fully implemented (see #192) sparse alternatives sets so that duplicate alternatives are dropped (and pick_count recorded) when the sample is initially created, so that logsums are only computed for unique alternatives, and the final interaction simulate only considers unique alternatives. This speeds up logsum computation and final interaction simulate runtimes considerably.
Tested the full scale example (all zones, all hhs, all skims) but ran into a bug that Jeff fixed. We also experimented with different chunk sizes to get better runtimes while not running into memory problems.
Started updating the documentation for the logsums revision and then will merge the pull request once done. I'll pull from the design notes. @Ben to finish.
Some discussion about whether we should do multi-processing/threading now instead of later, but we decided to continue with the workplan we have since we're building the software with chunking, which is the foundation for multi-processing/threading.
@Lisa, Wu to check on funding for FY2018; SFCTA, ARC are a go; PSRC will wait until 7/2018
Once the scope is final, @Bill will send the FY2018 invoice to the partners
@Joe, Wu talked with MWCOG and they are interested in joining, but probably not for a little while. They are working through a model update strategic plan which has an ABM development project in around 3 years, and then 2 years for ABM deployment. This works well with the ActivitySim project since the platform will be ready for prime time by then.
@Bill, Guy will share ActivitySim at the AMPO travel modeling working group meeting in DC on 9/25. They can use the Planning Apps slides as a starting point.
I'll starting planning two TRB meetups again this year - an informal get together and a scoping meeting