2025-12-18 Engineering Team

[joecastiglione]

Agenda

Review of Status

Summary of Progress this week

• Repair CDAP estimation activitysim#1026 bug fixed PR has been reviewed, minor changes before merge
• PSRC identified error in sharrow/legacy operation of trip destination model, 'utility not aligned' error when running sharrow in test mode using uv, but not in conda using same version of Activitysim. activitysim#1022, fix is implemented but needs testing
• Some effort investigating unexplained crashes in Sync production v15.3.1 + Sharrow sandag-abm3-example#39, no diagnosis yet but strong suspicion of numba memory problems due to inability to reliably reproduce the crash

[jpn--]

ActivitySim Development Meeting Notes

Meeting Overview

Agenda

• Review progress on recent bug fixes and pull requests
• Discuss memory issues with large-scale Sharrow models
• Update on shadow pricing zones work
• Plan for upcoming holiday schedule and Phase 11C tasks

Participants

• Joe Castiglione (@joecastiglione), Zephyr Foundation
• Jeff Newman (@jpn--), Driftless
• David Hensle (@dhensle), RSG
• Stefan Coe (@stefancoe), PSRC
• Joe Flood (@JoeJimFlood), CS

Duration

Approximately 17 minutes

Executive Summary

Jeff Newman reported progress on fixing the CDEP estimation bug and resolving a data frame matrix bug affecting PSRC's full-scale model with external zones. The team is investigating memory-related crashes in large-scale Sharrow models, particularly with SANDAG and MWCOG. RSG is close to completing their pull request on shadow pricing zones reopening, which represents their final outstanding task from Phase 11B.

Meeting Notes

Bug Fixes and Pull Requests

Jeff Newman provided updates on several technical fixes. The CDAP estimation problem has been resolved, with David Hensle reviewing and approving the PR with minor changes needed before merging. Jeff identified and fixed a bug in the data frame matrix code used in trip destinations that affected models with external zones. The issue occurred when running in test mode with both Sharrow and legacy code simultaneously, as they process zone gaps differently - Sharrow reindexes and compresses arrays while legacy code maintains gaps. A new pull request has been opened with the proposed fix, and Jeff plans to write a smaller test dataset to verify the solution more efficiently.

Memory Issues with Large-Scale Models

Jeff reported ongoing investigation into crashes affecting full-scale Sharrow models, particularly SANDAG. The crashes appear to be memory-related issues on Windows systems, as Jeff cannot reproduce them on his Mac laptop. The crashes are hard failures where the operating system terminates the process without Python or Numba being able to report errors. Joe Flood noted experiencing similar mysterious crashes during testing. Jeff believes the issues are size-related, affecting large models like SANDAG and MWCOG. David Hensle suggested the crashes might be related to memory profile traces from Sharrow not showing a constant baseline. Jeff acknowledged this connection and is working to identify proper memory allocation handling.

Shadow Pricing Zones Work

David Hensle reported RSG is close to opening their final pull request on shadow pricing zones reopening. They have replicated the issue with a small example and fixed the error. The main challenge is creating a self-contained unit test that doesn't require excessive additional files, as the functionality involves many calculations including workplace location specs, shadow pricing with land use files, and log sum calculations. The team expects to have the PR ready for review by mid-next week, which would complete RSG's outstanding Phase 11B tasks.

Settings File Issue

Jeff Newman and David Hensle briefly discussed an ongoing issue with a pull request intended to make index names uppercase instead of lowercase. The seemingly simple change has proven more complicated than expected, as changing capitalization in test files isn't working cleanly. They will continue working on this offline.

Phase 11C Planning

Joe Castiglione is finalizing the approach for Phase 11C and preparing to send communication to consultant teams. He shared a draft with the consortium members and is awaiting final approval before distributing it broadly. Joe emphasized the importance of maintaining momentum as RSG completes their Phase 11B tasks.

Holiday Schedule

The team confirmed cancellation of meetings during the Christmas and New Year's weeks. Meetings will resume the first week of January on the 6th and 8th. Meetings on January 13th and 15th during TRB week will remain scheduled, as most team members aren't attending the conference. Jeff noted the three-week gap is acceptable given the clear pathway to completing existing open tasks.

Action Items

• Jeff Newman: Merge CDAP estimation fix PR after addressing minor review comments
• Jeff Newman: Complete test for PSRC external zones bug fix and verify the solution works
• Jeff Newman: Continue investigating memory issues affecting large-scale Sharrow models (SANDAG, MWCOG)
• David Hensle/RSG: Complete pull request for shadow pricing zones reopening by mid-next week
• Jeff Newman & David Hensle: Continue working offline on settings file capitalization issue
• Stefan Coe: Test PSRC external zones fix once available
• Joe Castiglione: Finalize Phase 11C approach with consortium and send communication to consultant teams within 1-2 days