2026-07-09 Engineering Team

[jpn--]

Agenda

• Contracting Status
• Task Status / Coordination
• Open pull request follow-ups
• Performance testing dataset development

[bwentl]

@jpn-- If time permits, I would like to share a problem I encountered recently when trying to install both activitysim and populationsim in the same python environment with uv.

Here is the pyproject.toml & issues I faced:

[project]
name = "asim-py312"
version = "0.1.0"
description = "Test environment example"
readme = "README.md"
requires-python = ">=3.12"
dependencies = [
    "activitysim @ git+https://github.com/ActivitySim/activitysim.git@main",
    "populationsim @ git+https://github.com/ActivitySim/populationsim.git@master",
]

I ran uv sync, then I get this error:

  × No solution found when resolving dependencies:
  ╰─▶ Because only activitysim==1.5.2.dev19+g3bcc811bc is available and activitysim==1.5.2.dev19+g3bcc811bc depends on numpy>=2.0,<3, we can conclude that
      all versions of activitysim depend on numpy>=2.0,<3.
      And because populationsim==0.10.0 depends on numpy>=1.16.1,<2 and only populationsim==0.10.0 is available, we can conclude that all versions of
      activitysim and all versions of populationsim are incompatible.
      And because your project depends on activitysim and populationsim, we can conclude that your project's requirements are unsatisfiable.

[jpn--]

ActivitySim Consortium Meeting Notes

Participants

• Joe Castiglione (@joecastiglione) — Zephyr Foundation
• Jeff Newman (@jpn--) — Driftless
• Jan Zill (@janzill) — Outer Loop
• David Hensle (@dhensle) — RSG
• Sijia Wang (@i-am-sijia) — WSP
• Joe Flood (@JoeJimFlood) — CS
• Bo Wen (@bwentl) — TransLink
• Nick Fournier (@nick-fournier) — MTC

Duration

Approximately 30 minutes

Executive Summary

Joe Castiglione provided an update on contracting, noting the parent contract runs through 2027 and that an RFQ to reprocure the consortium bench is expected to be released around late July/early August, with firms identified by early October. The group touched on task status, including an upcoming calibration presentation from Jan Zill and David Hensle's teams, a reminder for feedback on Sijia Wang's explicit telecommute work, and background on how value-of-time heterogeneity relates to disaggregate accessibility ahead of Joel Freedman's upcoming presentation. Jeff Newman led a discussion on open pull requests, focusing on the stable-sorting-on-escort-tours PR, where he requested concrete evidence before adopting stable sorting over deterministic ordering. The meeting closed with an update on a NumPy version conflict between PopulationSim and ActivitySim, which Nick Fournier volunteered to investigate, and a brief status check with Joe Flood on the performance-testable data set effort.

Meeting Notes

Contracting and RFQ

Jeff Newman asked for clarification on the contracting situation, noting that individual task deadlines were set for June 30th while the parent contract runs through June 2027. Joe Castiglione confirmed there is no issue with task orders extending past their nominal end dates, and that no formal amendment is needed as long as scope and budget remain unchanged — he can send a confirming email if any firms want it in writing. He does not anticipate new "Phase 11" budgeted task orders at this time, despite some remaining funds, and suggested the group focus on bringing current tasks closer to completion rather than picking up new small tasks.

On the broader contracting front, Joe Castiglione explained that the consortium's bench was last competitively procured about five years ago, and agency procurement rules generally require a new competitive process roughly every five years. A draft RFQ has been circulated to agencies for feedback, with the intent to release it around late July/early August. If the schedule holds, firms for the bench would be identified by early October, with new contracts issued at that point superseding existing ones. Unlike the last procurement round, this RFQ will not include a set of conceptual response tasks; instead it will identify priority areas of interest (partly drawn from the current roadmap, partly new) and ask firms to describe their approach to those areas.

Task Status and Coordination

Jan Zill and David Hensle noted their teams are finalizing a EET presentation for the following Tuesday. Joe Castiglione asked whether it would help to share materials a day ahead so others could get oriented, but Jan explained the technical detail work is often finished close to the deadline; he'll aim to summarize the most important findings and provide brief background/context slides where possible rather than a full recap of prior reports.

Joe Castiglione flagged that feedback on Sijia Wang's explicit telecommute work was due roughly two weeks after her presentation, and said he would send a reminder to the group with a new deadline of the following Tuesday. Joel Freedman is scheduled to present telecommute frequency model updates the following week (the 21st); David Hensle confirmed Joel had already circulated a related email to some team members, which Joe Castiglione had seen but not yet fully reviewed.

David Hensle gave a brief primer on how a log-normal time-factor distribution is used to apply heterogeneity to value of time, and its implications for disaggregate accessibility and estimation — this should be turned off in accessibility/estimation runs and on in production runs, with more detail to come in Joel's presentation. Sijia Wang noted this relates to her explicit telecommute estimation work, where the random component of the mode choice log sum is set to one (already reflected in her presentation). David Hensle confirmed prior estimation results should remain valid, just with additional "fuzziness" introduced by the sampled time factor, and suggested an audit of which agency models use this approach to value-of-time heterogeneity. Jeff Newman noted more broadly that the EET (explicit error term) work has prompted exploration of several tangential issues in ActivitySim's random number handling that are not directly related to EET but surfaced as a result of that work.

Joe Castiglione also raised two smaller items: he plans to follow up with David Hensle to update some visualization maps (from earlier Phase 10 EET work) and needs to track down the relevant data request email, and he plans to check in with Bhargava Sana about SANDAG's review and summary of the shadow pricing issue, noting SANDAG has an open pull request in their own ABM repository related to that topic.

Open Pull Request Follow-ups

Jeff Newman noted that little progress has been made on most open pull requests beyond the stable-sorting-on-escort-tours PR, which has seen useful discussion about reformatting. He explained he left a comment requesting concrete evidence of a scenario where deterministic sorting fails and stable sorting is required, rather than adopting stable sorting as a precaution — his concern is that relying on stable sorting simply shifts the burden of deterministic ordering further upstream rather than resolving it. Joe Castiglione agreed and expressed hope that RSG or others involved would respond with either supporting evidence or a resolution.

Jan Zill noted that the main branch has moved ahead and that he has addressed some LLM code review comments and synced a related pull request in the Outer Loop repository with the current main branch; he asked Jeff not to make separate updates to the EET branch in the meantime, as he plans to merge additional updates into the EET branch shortly.

Performance Testable Data Set

Joe Flood reported partial progress on a performance-testable data set based on the Oceanside/Vista/Carlsbad area, which has more zones than the prior data set and includes a light rail line. He used a percentage of households to try to target a runtime of 20–25 minutes, but the run crashed early on; he set the work aside to revisit later. Jeff Newman asked him to return to it in the coming week or two.

PopulationSim/ActivitySim Dependency Conflict

Bo Wen flagged a dependency conflict encountered while setting up an environment containing both PopulationSim and ActivitySim. Jeff Newman explained that the latest ActivitySim release uses NumPy 2, while PopulationSim is still pinned to NumPy 1; the long-term fix is updating PopulationSim to support NumPy 2. David Hensle noted the required ActivitySim-side changes were relatively minor (mostly related to pinned dependencies such as Pandera), and since PopulationSim has fewer dependencies overall, the update may be straightforward. Nick Fournier volunteered to take a first pass at the PopulationSim update, noting he had already stripped out many of PopulationSim's dependencies previously. Nick also mentioned an unrelated fix related to a lingering issue for Joe Flood, caused by a Linux/Windows compatibility quirk.

Action Items

• Joe Castiglione: send a reminder to the group for feedback on Sijia Wang's explicit telecommute work, due next Tuesday
• Joe Castiglione: follow up with Bhargava Sana on SANDAG's review and summary of the shadow pricing issue
• Joe Castiglione: locate the prior data request email and coordinate with David Hensle on updating visualization maps
• Joe Castiglione and Jeff Newman: coordinate offline on rescheduling meeting times given Jeff's recurring conflicts
• Team: consider an audit of agency models using the time-factor approach to value-of-time heterogeneity
• Nick Fournier: take a first pass at resolving the NumPy 2 compatibility issue in PopulationSim
• Joe Flood: continue debugging the performance-testable data set crash, targeting the next week or two