Calculate highway reliability in warmstart and 1st global iteration only

[i-am-sijia]

What existing problem does the pull request solve and why should we include it?

This PR address #146

The updated reliability design:

• Reliability continues to be included in the VDFs. It has two parts: a static part and a LOS-based part. This is same as before.
• When running reliability, it continues to require two assignments. After the first assignment, reliability is written out as a link attribute, and gets skimmed in the second assignment. This is same as before.
• Implement a switch in the model config for reliability, so that users can choose to turn it on or off.
• If turned on, reliability is run in global iteration 0 (warmstart assignment) and 1 only. Assignment is run twice for glob iter 0 and 1, reliability is calculated and updated as a link-level attribute, and reliability skim is saved in the emmbank. Assignment is run only once for glob iter 2 and 3, the reliability skim does not get updated.
• If turned off, assignment is run once in all glob iter. Reliability is included in the VDFs, but not saved out as a link attribute, reliability skim will be skipped and be written out as 0.

What is the testing plan?

Should test run global iteration 1 and 2's highway assignment to make sure in global iter 1, assignment is run twice and highway reliability is calculated; and in glob iter 2, assignment is run once and highway reliability is not updated.

Demonstrate the code is solid by discussing how results are verified and covered by tests

• When reliability = true, check glob iter 1 calculated reliability, and glob iter 2 does not. Check the reliability skims from glob iter 1 and 2, to confirm they are the same.
• When reliability = false, check assignment is run once in both glob iter 1 and 2, and the reliability skims are all 0.

Questions

• When reliability = false, do we still want to keep the same VDF (with reliability in it)? If yes, then the travel time includes reliability, and the reliability in travel time is not consistent with the reliability skims.

Code formatting

Code should be PEP8 compliant before merging by running a package like black

• Code linted

Applicable Issues

Please do not create a Pull Request without creating an issue first.

Issues List

• closes 🚀 Feature: Change approach to creating reliability skims #146

[i-am-sijia]

@inrokevin I noticed @reliability == @static_rel for all links in the final assigned network which means the LOS based part of reliability was 0 in this step. I believe it's because here reliability_expr had get(1), but get(1) was not initialized. I removed the get(1) from the reliability expr. Please give a thumb-up if you agree with this approach.

[i-am-sijia]

To clarify, this is after I initialized put(1) in the BPR and Akcelik parts in the VDFs.

[inrokevin]

Agree, this is the simplest fix. I reviewed the changed expression and it looks correct. Thanks Sijia.