2026-01-29 Engineering Team

[jpn--]

Agenda

• Status of Remaining 11B Tasks
• New Issues/PRs:
  • Speed Up Nearest Zone Calculation in Disaggregate Accessibilities #1031
  • Status of explicit error terms branch
• Plan for discussion of testing on 2/12?

@ActivitySim/engineering

[jpn--]

ActivitySim Engineering Team Meeting Notes

Meeting Overview

Agenda

1. Status update on remaining Task 11B items
2. Discussion of new pull request from RSG
3. Explicit Error Terms branch updates
4. Planning for future testing discussion

Participants

• Joe Castiglione (@joecastiglione) - Zephyr Foundation
• Jeff Newman (@jpn--) - Driftless
• Sijia Wang (@i-am-sijia) - WSP
• David Hensle (@dhensle) - RSG
• Stefan Coe (@stefancoe) - PSRC
• John Gliebe (@johnpgliebe) - CS
• Craig Heither - CMAP

Duration

Approximately 41 minutes

Executive Summary

The team reviewed the status of remaining Task 11B items, with most nearing completion. Sijia Wang presented comprehensive updates to the skip households pull request, including new features for threshold setting, automatic tracing, and enhanced documentation. David Hensle discussed a new pull request that significantly improves performance for regions with large numbers of MAZs by vectorizing zone lookup operations. The team agreed to defer the testing strategy discussion to two weeks out and plans to close out Task 11B items before moving forward with Task 11C.

Meeting Notes

Task 11B Status Review

Jeff Newman provided an overview of remaining Task 11B tasks. Most outstanding work is either complete or in final review stages.

Skip Households Pull Request (WSP)

Sijia Wang reported completing all requested changes and adding several enhancements:

• Implemented threshold feature allowing users to set limits on number or weight of households that can be skipped before terminating the run
• Added summary output showing skipped households by model component at end of run
• Implemented automatic tracing for failed choices in simple simulate, eliminating need for manual trace_hh_id updates
• Noted that automatic tracing for interaction simulate is more difficult due to how interaction data frames are deleted; left this as manual step
• Updated unit tests to cover different threshold scenarios
• Added comprehensive documentation

Jeff Newman and David Hensle agreed to conduct thorough reviews of this pull request given its size and importance.

SANDAG Sharrow Issues

Jeff Newman reported ongoing work debugging Sharrow crashes on SANDAG servers. The model runs successfully on his laptop but fails on SANDAG servers. He modified test scripts to block threads to help isolate the problem. If current troubleshooting doesn't resolve the issue, he plans to work directly on SANDAG servers with Bhargava Sana.

Shadow Pricing Fix

Confirmed as completed, reviewed, and merged. However, this change caused CI test failures for prototype examples (one zone and two zone), which need reference validation updates due to minor changes in shadow pricing results.

UV Testing Deployment

CMAP reported successfully downloading UV and creating the environment but encountering errors when attempting to run ActivitySim. Craig Heither emailed Joe Flood about the issue. Joe Castiglione proposed giving it another week or two to resolve, then considering the task complete if all other agencies have successfully deployed UV, with CMAP issues addressed separately. Jeff Newman noted the problems may not be UV-specific but rather general model configuration issues.

New Performance Enhancement Pull Request (RSG)

David Hensle presented a pull request addressing performance issues PSRC encountered with their 66,000 MAZ system. Stefan Coe explained that the initialize_households step was taking approximately 25 minutes due to iterative zone lookups when assigning accessibility results to the synthetic population.

The fix includes two improvements:

• Vectorized the nearest zone lookup operation instead of iterating through each zone
• Optimized skim lookup to call once instead of repeatedly for every zone
• Performance improved from ~25 minutes to ~2 minutes

This issue hadn't appeared in other regions because they typically have around 20,000 MAZs rather than 65,000+. The operation is essentially n-squared, so the larger MAZ count caused disproportionate slowdown.

Explicit Error Terms Branch

Jeff Newman discussed a pull request from Matt Richards in New South Wales that merges recent main branch updates into the explicit error terms branch. The branch had gotten out of sync with ongoing main branch development. The pull request shows 300+ files changed, though the actual diff is smaller due to branch divergence.

Joe Castiglione suggested not waiting for Task 11C decisions and instead taking the opportunity to merge this work now. Jeff Newman agreed to review and work on this after completing higher priority items, and will communicate timeline expectations to Matt Richards. David Hensle noted the changes appear relatively minor, primarily involving estimation mode updates.

Task 11C Timeline

Joe Castiglione provided an update on Task 11C planning. The goal is to issue task orders within a month, with approximately:

• Two weeks for reviews of submitted proposals
• One week for negotiation
• One week to finalize
• Target: all task orders issued by end of February

Testing Strategy Discussion

Jeff Newman proposed scheduling the testing strategy discussion for two weeks from the meeting date (skipping the following week when he'll be on an Amtrak train). He will prepare a draft proposal for how testing should be structured, particularly for new additions, and how to move toward more robust unit testing rather than relying solely on large-scale model runs. The team will review and discuss his proposal at that meeting.

Action Items

• Jeff Newman & David Hensle: Conduct thorough review of skip households pull request from Sijia Wang
• Jeff Newman: Send Stefan Coe instructions for additional testing on utility alignment error
• Jeff Newman: Update reference validation for prototype examples (one zone and two zone) that are failing CI tests
• Jeff Newman: Continue debugging SANDAG Sharrow issues; coordinate with Bhargava Sana for server access if needed
• Joe Flood: Respond to Craig Heither's email regarding CMAP's UV deployment issues
• Joe Castiglione: Follow up on CMAP UV issues status in one week
• Jeff Newman: Review explicit error terms branch pull request from Matt Richards after completing higher priority items; communicate timeline to Matt
• Jeff Newman: Prepare draft testing strategy proposal for discussion in two weeks
• Joe Castiglione: Cancel next week's engineering team meeting