Project Meeting 2018.12.14

Parallelization

Completed optimization tweaks for Windows setup
90 min for the Windows 64 processor run (vs 60 min on Linux)
Ran into a bug in CDAP on Windows and Linux for the 128 processor setup
Maybe a slicing / chunking off-by-one indexing issue? Currently debugging
Should add production quality data typing to the future work program in order to reduce memory needs and improve performance

Now testing the shadowing pricing additions with 64 processors
Five iterations adds ~40 minutes of runtime currently
Saves the shadow prices from each iteration to the data store and can resume (i.e. warm start)
Working on verifying convergence, including spatially
Would be good to setup a place to share everyone's experience with shadow pricing configurations/use cases
There are lot of ideas around improved methods so this may be a good future work task

We want to setup and run TM1 on our server so we can compare results, @Lisa send 2013 RTP inputs and setup
We plan to work through the existing TM1 calibration/validation document high level submodel summaries
We're going to code up individual values-of-times now, before validation, since this was not done initially and will be important for validation
Will be done through annotate hhs and persons expressions
Maybe implement as a Jupyter notebook example/tutorial
Visualization of results with ActivityViz would be helpful

Both the tutorial/workshop and parallelization papers were accepted
Workshop will be an introduction to the initiative and the software
Maybe 2 to 3 hours in length, maybe some hands-on
Goals - demonstrate the real powerful tool and excite potential new team members
Would be good to add reporting/analysis tools as well for demonstration purposes
Let's revisit in a few months, probably after verification