Track Sprint 3 Configuration Changes

[DavidOry]

@i-am-sijia, @vivverma9

Let's use this branch to keep track of the configuration changes we are making in support of the sprint 3 runtime improvements.

• Config cleaning
• Warmstart updates
• Highway assignment relative gaps specific to global iteration
• Transit journey level changes
• Highway reliability changes
• Transit assignment relative gaps and max iterations specific to global iteration
• OpenPaths compatible

[i-am-sijia]

The config changes we are going to track here will require different PRs of tm2py. E.g., the journey level changes require (I suppose) BayAreaMetro/tm2py#153. The relative gap changes require BayAreaMetro/tm2py#156. We do not have a single version of code right now that can test these config as is. It's ok, just requires a bit more manual tracking, like when I use config from here, I will need to comment out the journey level pieces. If we separate out the config changes into different branches, we can test them with their respective tm2py PRs.

[DavidOry]

The config changes we are going to track here will require different PRs of tm2py. E.g., the journey level changes require (I suppose) BayAreaMetro/tm2py#153. The relative gap changes require BayAreaMetro/tm2py#156. We do not have a single version of code right now that can test these config as is. It's ok, just requires a bit more manual tracking, like when I use config from here, I will need to comment out the journey level pieces. If we separate out the config changes into different branches, we can test them with their respective tm2py PRs.

Right. I think the idea is once the all the PRs have been merged and we have a new release of tm2py, we'll want to have a record of the configuration file we use for the sprint 3 model run. But yes, this does not replace the testing of the individual PRs.

[i-am-sijia]

Ready for sprint 3 test run. Changed settings in sprint 3:

• Reliability is set to TRUE, which makes highway assignment run twice in glob iter 0 and 1, and once in glob iter 2 and 3. Reliability skims remain unchanged after glob iter 1.
• Highway assignment relative gap is set to 0.05 for glob iter 0-2, and 0.0005 for glob iter 3.
• Transit assignment is set to run 2 (instead of 10) max iterations for AM and PM in glob iter 1.
• Transit assignment journey_levels is using manual specification with 2 levels.
• num_processors is set to MAX-1 instead of a hardcoded int

[DavidOry]

Ready for sprint 3 test run. Changed settings in sprint 3:

• Reliability is set to TRUE, which makes highway assignment run twice in glob iter 0 and 1, and once in glob iter 2 and 3. Reliability skims remain unchanged after glob iter 1.
• Highway assignment relative gap is set to 0.05 for glob iter 0-2, and 0.0005 for glob iter 3.
• Transit assignment is set to run 2 (instead of 10) max iterations for AM and PM in glob iter 1.
• Transit assignment journey_levels is using manual specification with 2 levels.
• num_processors is set to MAX-1 instead of a hardcoded int

@i-am-sijia: please use 0.05 for the last highway assignment global iteration. And set max transit iterations to three for global iteration 2 and five for global iteration 3.

thanks.

[AshishKuls]

@DavidOry Why are we doing 0.05 relative gap for even the last global iteration. Did we not decide to use a tighter gap for final one?

[DavidOry]

@DavidOry Why are we doing 0.05 relative gap for even the last global iteration. Did we not decide to use a tighter gap for final one?

I think there are some use cases in which you would want a tighter assignment for the last global iteration, e.g., when comparing this run against another with a relatively modest roadway change. But there are lots of situations in which this would not be necessary, e.g., transit studies, comparing very different regional scenarios. I think it's useful to know the runtime for when a tight assignment is not needed.