Activity

1. joecastiglione

   added

   meetingMeeting notes.Meeting notes.

   on Aug 27, 2026
2. janzill commented on Sep 9, 2026

   

   janzill

   on Sep 9, 2026

   More actions

   ActivitySim Engineering Team Meeting Notes — September 3, 2026

   Meeting Overview

   Participants

   • Jan Zill (@janzill) — Outer Loop
   • Joe Castiglione (@joecastiglione) — Zephyr Foundation
   • Amir Samimi (@amsamimi) — Transport for NSW
   • Andrew Kay (@andkay) — Cambridge Systematics
   • Guy Rousseau (@guyrousseau) — ARC

   Duration

   51 minutes

   Executive Summary

   The team reviewed work approaching completion at the end of Phase 11 and supported making a new ActivitySim release soon after the park-and-ride pull request is ready. The exact version, whether to label it as a beta and a longer-term release-frequency policy were left for the Product and Community Team. Participants noted both the need for agencies to access unreleased features through pinned package versions and the testing risk created by accumulating many significant changes between releases.

   Several reviews still require clearer acceptance criteria or documentation. The calibration draft needs searchable, user-facing guidance in addition to its developer README. The faster random-number implementation needs agreed tests for performance, memory and expected output differences. Review of the Application and Analysis Guide exposed an unresolved access problem: Google Docs is blocked by some agencies, while the earlier GitHub review workflow reportedly made comments difficult to track as the document changed.

   Meeting Notes

   Calibration Tooling

   • #1107 Calibration was available as a draft, but its estimation notebook and estimation unit tests were failing. Detailed review was deferred until David could attend, although participants were encouraged to record high-level feedback on the pull request.
   • The README under model_core was considered developer-oriented and insufficient as the only documentation. The team requested searchable Sphinx documentation, an example setup and concise guidance alongside the YAML configuration that users interact with.

   Faster Random Numbers

   • Jan had completed most of his review of #1103 Faster random numbers, with one or two technical questions remaining before returning comments to Jeff.
   • The legacy random-number generator remains the default and is intended to preserve existing results. Enabling the new generator will change individual results, but the aggregate differences should resemble seed-to-seed variation rather than systematic drift.
   • David had previously offered to add large-scale before-and-after runtime and memory profiles. Amir also requested that the expected result comparison and acceptance criteria be stated explicitly by the developer before other reviewers add tests.

   Application and Analysis Guide Review

   • Review of #1086 Application and Analysis Guide had moved to a shared Google document after line-specific GitHub comments reportedly became disconnected from the changing source lines. The review period could extend by another week to allow substantive agency feedback.
   • Amir reported that Transport for NSW cannot access Google Docs from its managed environment. He favoured keeping review discussion with the Markdown source in GitHub so comments, replies and resolving commits remain traceable.
   • Joe Castiglione and Andrew had not yet established whether the earlier GitHub difficulty was inherent to pull-request review or specific to the document's volume of comments and incremental edits. Andrew agreed to clarify the problem with Joe Flood; no review platform was selected during the meeting.

   Remaining ActivitySim Pull Requests

   • #1083 Shadow-pricing segmentation and #1062 Summarise memory cleanup still needed unit tests. Their existing owners were absent, so no further decision was made.
   • Changes to #1001 Park-and-ride lot choice and capacity had been approved, and the pull request was expected to merge once the remaining continuous-integration tests passed.
   • Further engineering discussion of #1106 Automatic adaptive chunking was deferred until the full technical team was present and the Product and Community Team had gathered agency perspectives. Transport for NSW had used adaptive chunking before moving to explicit chunking; the open product question was which agencies still use adaptive chunking and what user value automatic memory-based configuration would provide.

   Merged Skim Support and Continuous Integration

   • #1101 Parquet skim support had merged with the agreed scope of putting the capability into production before undertaking further optimisation.
   • ActivitySim's main-branch checks then failed because publishing approximately 100 MB of test artifacts after every merge had filled TestPyPI's 10 GB project allowance. #1109 Stop publishing to TestPyPI removed that non-essential deployment step while retaining release publication to PyPI.
   • A merge conflict affecting #1036 Unused-skim loading was caught by continuous integration, fixed and merged.

   OMX and Zarr Skim Precedence

   • #1108 Zarr/OMX skim precedence reported inconsistent behaviour when both formats are present. An OMX file was still read before a valid Zarr store and therefore was not optional, while behaviour also differed between sharrow shared-memory loading and the non-sharrow path.
   • The external contributor offered a narrow patch, and potential additional changes to sharrow. The team suggested that Jeff assess it because related, not-yet-open work was moving some OMX functionality to lower-level HDF5 libraries for read/write performance. Ownership was not confirmed during the meeting.

   PopulationSim Updates

   • #219 Dependency modernisation and cross-platform stability had been reviewed favourably. One remaining comment concerned moving a check out of a workflow and into a test; resolving it was expected to allow the pull request to merge and unblock the intermediate-geography work.
   • Investigation of #221 Household-weight hard constraints had exposed additional possible problems, including one that might require more substantial work. Amir had not yet run the requested tests but retained them as a side task.

   Release Timing and Policy

   • Amir requested a release from the main branch because agencies that pin ActivitySim package versions cannot adopt the many useful features merged since the previous release. He proposed that a beta could make limited testing explicit while allowing agencies to provide feedback sooner.
   • Joe Castiglione supported releasing soon but was concerned that many significant changes had accumulated, including externally led work that the consortium had followed but not managed directly. He proposed a more explicit minor-release policy so changes do not continue to stack up between releases.
   • The team supported preparing a release after #1001 was ready. The Product and Community Team was left to decide the version, beta status and longer-term release cadence; the draft calibration work was not considered a reason to delay this release.

   Modeling Mobility Abstracts

   • The team revisited whether to submit a collective Modeling Mobility proposal. Options included a broad, workshop-style project update and narrower presentations on particular improvements, testing or new features. No choice had been made in the earlier discussion.
   • Abstracts were understood to be due on September 19. The target was to decide what, if anything, to submit by the September 10 engineering meeting, while recognising that contributors remained free to submit independent proposals.
   • Explicit error terms and the random-number work were identified as possible technical topics. Jan said the collective plan would determine whether he instead pursued an explicit-error-terms submission separately.

   Action Items

   • Jan Zill (@janzill) will add the request for user-facing calibration guidance to #1107.
   • Jan Zill (@janzill) will complete his technical review of #1103 and return comments to Jeff Newman (@jpn--).
   • Andrew Kay (@andkay) will ask Joe Flood (@JoeJimFlood) what specifically failed in the earlier GitHub review workflow for #1086.
   • Amir Samimi (@amsamimi) will run the outstanding tests for #221.
   • Joe Castiglione (@joecastiglione) will prepare and circulate notes from the September 1 Modeling Mobility discussion before the next engineering meeting.
3. janzill

   closed this as completedon Oct 1, 2026