Compare VP- and TU-based stop arrivals data

[mrtopsyt]

Changes

• Added notebook and sql queries to compare stop arrivals generated from Vehicle Positions (VP) in the speeds pipeline and from Trip Updates (TU) in the fct_stop_times_metrics table on dates where both are available.

Summary of Findings

• TU-based data is generally available for a greater number of trips than VP-based data
• For different agencies checked (SamTrans, SFMTA, SBMTD, SacRT), 83-90% of stop arrivals available for both agencies matched within 60 sec. Lower levels of alignment were anecdotally associated with MTC Regional Feeds.
• For SamTrans specifically, TU stop arrivals snapped to within an integer multiple of 60 seconds from a schedule, which was not observed for VP.
• For all agencies checked, TU data was associated with a 5-10x greater incidence of non-monotonic arrivals, however the magnitude of these non-monotonic cases was smaller. Note that the VP data is subject to data cleaning steps designed to decrease non-monotonicity, whereas the TU-based data is not

[mrtopsyt]

@hhmckay @edasmalchi, this is the notebook I presented to you! Merging because it's all new files, but wanted to ping you in case you want to take a look

[github-actions]

nbviewer URLs for impacted notebooks:

• gtfs_rt_comparison/compare_vp_tu.ipynb