Activity

1. joecastiglione

   added

   meetingMeeting notes.Meeting notes.

   on Sep 25, 2026
2. janzill commented on Oct 2, 2026

   

   janzill

   on Oct 2, 2026

   More actions

   ActivitySim Engineering Team Meeting Notes — October 1, 2026

   Meeting Overview

   Participants

   • Jan Zill (@janzill) — Outer Loop
   • Joe Castiglione (@joecastiglione) — Zephyr Foundation
   • Nick Fournier (@nick-fournier) — MTC
   • David Hensle (@dhensle) — RSG
   • Sijia Wang (@i-am-sijia) — WSP
   • Andrew Kay (@andkay) — Cambridge Systematics
   • Amir Samimi (@amsamimi) — Transport for NSW

   Duration

   54 minutes

   Executive Summary

   PopulationSim's stabilization pull request was ready for David's final check, and the group favored testing its supported Python versions across Linux, Windows, and macOS. For ActivitySim, Jan proposed making Python 3.11–3.13 explicit in continuous integration; Python 3.14 remained blocked by a Sharrow compatibility issue. Several ActivitySim pull requests were close to review or merge, with example-model validation, documentation, and further testing still outstanding.

   The team began a broader discussion of release policy without settling a cadence or required test threshold. Andrew suggested a target cadence for feature releases, while Jan favored prompt, small patch releases; Joe urged a structured policy that also addresses model-scale, statistical, and performance validation. Joe C. took responsibility for submitting the Modeling Mobility session proposal by its October 2 deadline. Amir proposed technical panels for IATBR 2027 in Sydney.

   Meeting Notes

   PopulationSim Stabilization and Python Testing

   • Nick had addressed David's review comment on PopulationSim #219 Stabilization, removed redundant test coverage, and added Python 3.14 to the Ubuntu test matrix. David said the changes looked good and would give the pull request a final check before merging it if tests passed. The merge would unblock the intermediate-geographies work.
   • Jan proposed extending the matrix to Python 3.10–3.14 on Windows and macOS as well as Ubuntu. Nick supported testing the full combination so that stated support is checked on each operating system.

   ActivitySim Python Support and Windows Installer

   • Jan had run ActivitySim unit tests locally on Linux with Python 3.11–3.13 and proposed adding these versions to continuous integration. A Windows Python 3.13 failure and a dependency-resolution issue still needed investigation. Python 3.14 was blocked by Sharrow using an abstract-syntax-tree API that is no longer available in that version.
   • Jan said the repository's Python setting had made one version the routine development choice, while the package metadata permitted other versions. Explicit tests would make support for versions already used in agency projects clearer. Jan planned to investigate the Windows failure and then open a pull request for review at the next meeting.
   • The group revisited its earlier decision to remove the Windows installer, which depends on an outdated conda-forge package and Python 3.10. Jan offered to open an issue and pull request, allow time for feedback, and then proceed with removal.

   ActivitySim Pull Requests

   • #1083 Simulation-based shadow-pricing segment defaults was ready for review, but regression tests for two example models failed because the Sharrow name fix changed work location-choice validation results. Sijia will open pull requests to update both example-model targets. Jan will check whether Jeff has reviewed the core change and review it if needed, so the related changes can be merged together.
   • Jeff was addressing Jan's review comments on #1103 New RNG. Jan had approved the change but would check whether anything remained outstanding.
   • David and Jan continued reviewing #1110 Reduce peak memory. Jan proposed applying the extra processing only in reduced-memory mode to address runtime concerns. The developing Boston model appeared to work, and MTC results were available, but Jan wanted a few code and documentation changes before merge. He noted that expressions depending on an entire population could yield different results when data are processed in chunks; David mentioned that multiprocessing already limits most workers to their assigned data unless tables are explicitly shared.
   • #1119 Extension loading addresses #1118: in multiprocessing mode, an extension may fail to load unless its folder is in the directory from which ActivitySim is invoked. Sijia will review and test the fix against the Boston work, where this problem was found.
   • For #1120 Run manifest, Jan supported the idea of a documented, machine-readable account of the components actually run, rather than relying on information scattered across settings, logs, and other outputs. The group deferred detailed discussion until Jeff could provide more context at the next meeting.

   Testing and Release Policy

   • Jan described the manual work of running full-scale SANDAG and MTC models and comparing results before a release. The team had not decided whether every release needs those runs, exact output matches, or statistical acceptance criteria. Jan suggested considering Joe Flood's smaller SANDAG test model as a possible way to broaden routine coverage at lower cost.
   • Andrew said the release-policy question also concerned timing: completed features can wait while other work remains in progress. He suggested a target, such as quarterly feature releases, while allowing exceptions. Jan favored releasing small bug fixes promptly as patches. No cadence was agreed.
   • Joe C. called for a structured engineering-led discussion that combines release timing with test coverage, performance comparisons, acceptance thresholds, possible beta releases, and agencies' own release-adoption policies. Jan agreed that the team should prepare a document setting out options and trade-offs; no owner or deadline was assigned.

   Modeling Mobility and IATBR 2027

   • David confirmed that he had not submitted the proposed Modeling Mobility ActivitySim session. Joe C. will turn the existing outline into a submission from the consortium and submit it by the October 2 deadline. Joe C. also offered to ask the conference organizers whether online participation would be available; he expected that cost could make it difficult.
   • Amir proposed using IATBR 2027 in Sydney to convene researchers, consultants, and public agencies around technical questions in activity-based modeling. Possible panels could examine scheduling, destination choice, and newer model formulations. Jan and Joe C. supported developing the idea and discussing how practical model choices relate to academic approaches, but the group did not assign a submission task.

   Action Items

   • David (@dhensle) will give PopulationSim #219 a final review and merge it if the tests pass.
   • Jan (@janzill) will investigate the ActivitySim Python 3.13 Windows failure and open a pull request for an expanded Python test matrix.
   • Jan (@janzill) will open an issue and pull request to remove the Windows installer, allowing time for feedback before removal.
   • Sijia (@i-am-sijia) will open pull requests updating validation targets in the two example models affected by #1083.
   • Jan (@janzill) will check whether Jeff has reviewed #1083 and review it if needed.
   • Jan (@janzill) will add his remaining review comments to #1110.
   • Sijia (@i-am-sijia) will review and test #1119 against the Boston model.
   • Joe C. (@joecastiglione) will prepare and submit the consortium's Modeling Mobility session proposal by October 2.
   • Joe C. (@joecastiglione) will ask the Modeling Mobility organizers whether online participation will be available.