2025-11-06 Engineering Team

[joecastiglione]

@ActivitySim/engineering

Agenda

• Review of Current Work Status
  • UV Documentation, any comments? esp. on UV for scripts
  • SANDAG example with Sharrow, needs Windows performance testing on this
  • Reproducible Random on trip scheduling choice -- testing code
• Any more engineering input on possible 11C tasks?
• Park and Ride Implementation Discussion

[jpn--]

ActivitySim Engineering Team Meeting Notes

Meeting Overview

Agenda

• Review current work status for Phase 11
• Discuss scoping needs for task 11C
• Park and ride implementation discussion (if time permits)

Participants

• Jeff Newman (@jpn--), Driftless
• Joe Castiglione (@joecastiglione), Zephyr Foundation
• David Hensle (@dhensle), RSG
• Sijia Wang (@i-am-sijia), WSP
• Stefan Coe (@stefancoe), PSRC
• Andrew Kay (@andkay), CS
• Ray Ngo, Metropolitan Washington Council of Governments

Duration

Approximately 60 minutes

Executive Summary

The engineering team reviewed the status of Phase 11 work items, including progress on reproducible random numbers, UV migration documentation, and SANDAG model updates. Key decisions included setting a Monday deadline for UV documentation review and merge, postponing input data updates for the SANDAG example model, and discussing park and ride implementation details. PopulationSim 0.10 testing by MWCOG confirmed that randomness issues have been resolved.

Meeting Notes

Phase 11 Project Status Review

Jeff Newman organized all Phase 11 work into a consolidated project board, distinguishing between contracted tasks with specific budgets and general support items.

Reproducible Random Numbers in Trip Scheduling

• RSG completed the work; Jeff Newman reviewed and found the solution effective
• Jeff created a sub-pull request to improve testing methodology, focusing on unit testing of specific components rather than config file validation
• Requested David Hensle review the testing approach as a prototype for future component-level unit tests
• Goal is to establish documentation standards for unit testing components versus whole model systems

Zone Reopening with Shadow Price Simulation

• No progress yet; RSG has been working on other priorities

Trace Files Hash Consistency

• Joe F. flagged Jeff for coordination; Jeff needs to follow up

Command Prompt Logging Cleanup

• Original pull request merged successfully for legacy models
• However, Sharrow mode still produces excessive logging output
• Jeff asked CS to check if budget remains in the task for additional cleanup work

Alternatives Files Formatting

• RSG developed a solution requiring testing with SANDAG estimation data
• Solution enforces lowercase alias names consistently across all alternative files
• Includes deprecation warnings rather than breaking changes to allow gradual migration
• Team discussed need to eventually remove deprecated features that have been unsupported for 6-7 years

SANDAG Example Model Updates

Sharrow Consistency

• Jeff completed work to make SANDAG example model work with Sharrow
• Testing on Mac shows performance improvements
• Critical requirement: Need Windows server testing at SANDAG to confirm performance gains before merging
• Jeff will tag Bhargava Sana to request SANDAG testing

Input Data Updates Discussion

• Bhargava Sana requested updating SANDAG example model inputs to match latest production model data
• Jeff expressed concerns about maintaining up-to-date inputs:
  • Every input change breaks all regression tests, requiring regeneration
  • Manual review needed to verify no errors introduced
  • Example model serves consortium testing needs, not production modeling
  • Stable inputs more valuable than currency for testing purposes
• Joe Castiglione suggested SANDAG could generate cropped dataset and results for testing
• Team agreed to discuss offline with SANDAG to minimize consortium resource burden

UV Migration and Documentation

Current Status

• Jeff and others reviewed the UV pull request
• David Hensle added suggestions for Sijia Wang to consider
• Jeff discovered UV can use script-level lock files for isolated environments in SANDAG example scripts

UV Script-Level Environments Discussion

• Jeff demonstrated UV's ability to create isolated environments at individual script level
• Sijia Wang noted this feature was added for machine learning prototyping but production use should employ project-level UV setup
• David Hensle raised concerns about maintenance burden of multiple script-specific dependencies
• Team decided to backburner this approach for now

Timeline and Next Steps

• Deadline set for Monday: Sijia Wang to address David Hensle's comments
• All reviewers should complete reviews by Monday morning
• Jeff will merge and create version 5.1 mini-release
• Documentation will be published to web for next round of agency testing
• Second round will include agencies not involved in first round (ARC, TfNSW) plus additional staff at agencies who participated previously

Other Work Items

Global Option to Skip Households

• Sijia Wang making small progress; will increase effort

Reading Skims in Parquet Format

• Jeff messaged Amarin about compression results and skim format details
• Awaiting agency feedback before declaring prototype format

Population Outputs Randomness

• Jeff messaged MWCOG to test with PopulationSim 0.10
• Ray Ngo reported: Testing completed successfully, randomness issue resolved
• Integer timeout value confirmed to have no impact

Documentation Front Page Updates

• Jeff opened pull request adding cards for consortium examples and member agency models
• Included links to PSRC, ARC, Metropolitan Council, Oregon DOT
• Requested agencies provide links if they want models featured

Park and Ride Implementation

David Hensle presented detailed park and ride model design for 15-minute discussion.

Conceptual Framework

• Park and ride lot choice runs for every tour
• Tour mode choice determines who actually uses park and ride
• Model counts capacities at each lot
• Overcapacity lots trigger re-simulation of latest arrivals
• Process iterates until capacity constraints satisfied

Utility Specifications

• Driving utilities: Origin to lot skim (SOV time at medium VOT) for outbound and inbound
• Transit utilities: Lot to destination skims including first wait time, transfer time, walk time, vehicle time, penalties, and fares
• Size variable: Number of parking spaces (more spaces = higher attraction)
• Binary variable: Lot fullness status (full lots become unavailable)

Technical Implementation

• Creates two-pronged origin-lot-destination skim structure
• Consolidated skim wrapper functions into modular approach for cleaner code
• Choice set includes all zones with parking spaces > 0
• Multiple lots in same zone are summed

Feedback and Iteration Discussion

• Joe Castiglione asked about time-of-day feedback when lots fill up
• Current design iterates only with tour mode choice, not with upstream models like time of day
• David Hensle noted complexity of iterating with upstream models
• Joe clarified he was not suggesting iterating with upstream models, only that there should be a signal in the global feedback skims reflecting PNR impedance by ODT
• Jeff Newman suggested aggregate lot fullness data could feed back through global iterations without disaggregated tracking
• Proposed solution: Include utility term based on departure time compared to typical lot fill-up times
• Stefan Coe noted this represents significant improvement over current assignment-based approaches even without time-of-day feedback
• Sijia Wang described MTC's approach using capacitated transit assignment with feedback to demand model

Action Items

• David Hensle (@dhensle): Review Jeff Newman's sub-pull request on trip scheduling choice unit testing approach
• Andrew Kay (@andkay): Check with CS on remaining budget for Sharrow logging cleanup
• Jeff Newman (@jpn--):
  • Follow up with Joe F. on trace files hash coordination
  • Tag Bhargava Sana to request SANDAG Windows server performance testing
  • Discuss SANDAG input data updates offline with Bhargava Sana
  • Merge UV pull request and create version 5.1 release by Monday
• Sijia Wang (@i-am-sijia):
  • Address David Hensle's comments on UV pull request by Monday
  • Continue progress on global option to skip households
  • Follow up with ARC and TfNSW for second round UV testing
• All reviewers: Complete UV pull request reviews by Monday morning
• Jeff Newman / Joe Castiglione: Consider opening issue for removing deprecated features from 6-7 years ago
• David Hensle: Schedule park and ride presentation for broader Tuesday meeting to discuss outputs and results