Agenda
- Explicit Telecommute Implementation Kickoff
- Model Calibration Partial Automation: Design Kickoff
- Integration of EET-related development
- Poisson sampling validation tasks and resource requirements
- Testing strategy
- Documentation
- "Optionality" design concerns
- EET output analysis schedule
- ActivitySim Application & Analysis Guide Discussions
- RNG Testing
- Admin
- Phase 12 / FY27 agreements
Notes
Explicit Telecommute Implementation Kickoff
The telecommute implementation task kicked off last week, led by WSP. This phase covers the first two model components from the broader design developed in Phases 9–10: the telecommute arrangement model and the in-home work activity duration model.
One issue raised in the kickoff was that the in-home work activity duration distribution was slightly unexpected. WSP (Sijia) will dig into the survey data further, segmenting respondents into those who always work from home, those who episodically do, and those splitting work between home and outside. A known limitation is that U.S. household travel diary surveys capture in-home activities only superficially, which constrains the model's level of detail.
Interim check-ins with the consortium have been requested so members can review and guide WSP's model specification before it is finalized. Joe reached out to Sijia to schedule these.
2. Model Calibration Partial Automation: Design Kickoff
The design phase kicked off today in the consortium's earlier meeting, led by RSG (Joel Freedman and David Hensle). Joel briefly recapped: scope and motivation were presented, an initial design and implementation proposal was shared, and discussion followed on flexibility and convergence criteria. Consortium members are invited to comment on the slides and documentation once circulated.
TfNSW parallel effort: Amir noted that TfNSW has already developed an automated calibration tool internally, which RSG has access to. He asked Shaun to explore whether TfNSW can open-source the code so RSG can use it as a starting point. Shaun agreed to investigate. Even if the code cannot be shared publicly, RSG can review the configuration and design documentation. Amir and Joe expressed interest in TfNSW presenting their tool to the broader consortium once more robust testing is complete.
3. EET Integration Strategy
Jan Zill provided a detailed update on Explicit Error Terms (EET) integration. Key points:
- The main EET production-readiness work (testing, documentation, inconsistency fixes, runtime fix) is in a pull request awaiting review.
- VLC has implemented Poisson sampling, the most promising runtime alternative. It is close to runtime parity (~10% slower per demand model iteration) but offers significantly improved stability and fewer inconsistencies between base and scenario runs.
- Jan's recommendation: get Poisson sampling into production now as a pragmatic step. Estimated effort: ~25 hours base code + ~15 hours for validation and large-scale model runs. Joe has reached out to VLC to discuss whether it is possible to contractually engage then to do this work. Alternatively, Outer Loop could complete this work. Joe also needs to reach out to the consortium to discuss funding this from the contingency — no decision required from the group today.
- Open questions remain on whether to maintain multiple sampling methods going forward, and whether improved RNG work will bring current EET sampling to full runtime parity with Poisson.
- Jan is also investigating a "Best-K" sampling approach (drawing error terms and keeping only the top K options) discussed at last weeks engineering meeting, which could further reduce sampling/choice inconsistencies. He will report back at the next engineering meeting.
LUTI/scenario question (Paris): Paris raised how EET behaves when land use changes between scenarios, e.g., when integrating a LUTI model. VLC's final report will comment on this. Jan noted the inconsistency investigation partially addresses it via a single-iteration land-use-change test, but the broader multi-iteration question is open for future work.
4. EET Output Analysis Schedule
RSG will present RNG testing results next week; early results were described as encouraging. Outer Loop and RSG are each expected to report back on their investigation work towards the end of May, with a task deadline of end of June. Key decisions still pending: which code version to use, which features to include in testing, and which model to test against.
5. ActivitySim Application & Analysis Guide
Members are asked to continue reviewing and commenting on the five guide topic links in the agenda. Amir raised a concern that the network change guidance should focus on how network changes (e.g., introducing a new mode like bike-and-ride) are reflected in ActivitySim's demand model — not network coding in the traffic assignment model. He noted a comment he left two weeks ago may have been marked out of scope. Alex agreed this should be discussed further rather than resolved in the comments alone, and will flag it with Joe.
Amir also requested that any decision-making meetings on guide content be scheduled at Australia-friendly times. Alex committed to raising this with Joe.
6. Random Number Generation Testing
A report-out on RNG testing is scheduled for next week. Early indications from David suggest results are encouraging. This work is part of the broader EET effort led by RSG with support from Driftless andWSP, focused on replacing the legacy RNG components.
7. Admin — Phase 12 Agreements
Phase 12 (July 1 start) agreements are going out. The goal is to proceed with existing language and incorporate agency feedback in the 2027 cycle. Agencies with hard sticking points can raise them now.
- TfNSW: Signed the current agreement but had minor suggested clause changes. Shaun will circulate these to Joe for future consideration.
- Categorization question (Paris/Kate): Agencies have struggled to classify consortium membership in procurement systems. The practical approach being used by some (following Ohio's lead) is to classify it as a low-cost software purchase. Kate noted her agency secured a two-year sign-off, which should carry through this cycle.
Engineering meeting timing: Amir noted that engineering meetings are fixed at 4 AM Sydney time, unlike the executive meetings which rotate. He asked whether the engineering meeting could be shifted to alternate between Australia-friendly and US-friendly times. Alex committed to raising this with Joe.
Action Items
| # |
Action |
Owner |
| 1 |
Explore open-sourcing TfNSW calibration tool; keep Amir updated |
Shaun Tabone |
| 2 |
Schedule interim telecommute check-ins with Sijia (WSP) |
Joe Castiglione |
| 3 |
Follow up with VLC re: funding Poisson sampling production work |
Joe Castiglione |
| 4 |
Circulate TfNSW Phase 12 agreement clause suggestions to Joe |
Shaun Tabone |
| 5 |
Flag guidebook scope concern & Australian meeting time request to Joe |
Alex Bettinardi |
| 6 |
Continue reviewing and commenting on guide topic links |
All agencies |
| 7 |
Present RNG testing results |
RSG (next week) |
| 8 |
Investigate Best-K sampling; report at engineering meeting |
Jan Zill |
Notes prepared from transcript. Please flag any corrections.
Agenda
Notes
Explicit Telecommute Implementation Kickoff
The telecommute implementation task kicked off last week, led by WSP. This phase covers the first two model components from the broader design developed in Phases 9–10: the telecommute arrangement model and the in-home work activity duration model.
One issue raised in the kickoff was that the in-home work activity duration distribution was slightly unexpected. WSP (Sijia) will dig into the survey data further, segmenting respondents into those who always work from home, those who episodically do, and those splitting work between home and outside. A known limitation is that U.S. household travel diary surveys capture in-home activities only superficially, which constrains the model's level of detail.
Interim check-ins with the consortium have been requested so members can review and guide WSP's model specification before it is finalized. Joe reached out to Sijia to schedule these.
2. Model Calibration Partial Automation: Design Kickoff
The design phase kicked off today in the consortium's earlier meeting, led by RSG (Joel Freedman and David Hensle). Joel briefly recapped: scope and motivation were presented, an initial design and implementation proposal was shared, and discussion followed on flexibility and convergence criteria. Consortium members are invited to comment on the slides and documentation once circulated.
TfNSW parallel effort: Amir noted that TfNSW has already developed an automated calibration tool internally, which RSG has access to. He asked Shaun to explore whether TfNSW can open-source the code so RSG can use it as a starting point. Shaun agreed to investigate. Even if the code cannot be shared publicly, RSG can review the configuration and design documentation. Amir and Joe expressed interest in TfNSW presenting their tool to the broader consortium once more robust testing is complete.
3. EET Integration Strategy
Jan Zill provided a detailed update on Explicit Error Terms (EET) integration. Key points:
LUTI/scenario question (Paris): Paris raised how EET behaves when land use changes between scenarios, e.g., when integrating a LUTI model. VLC's final report will comment on this. Jan noted the inconsistency investigation partially addresses it via a single-iteration land-use-change test, but the broader multi-iteration question is open for future work.
4. EET Output Analysis Schedule
RSG will present RNG testing results next week; early results were described as encouraging. Outer Loop and RSG are each expected to report back on their investigation work towards the end of May, with a task deadline of end of June. Key decisions still pending: which code version to use, which features to include in testing, and which model to test against.
5. ActivitySim Application & Analysis Guide
Members are asked to continue reviewing and commenting on the five guide topic links in the agenda. Amir raised a concern that the network change guidance should focus on how network changes (e.g., introducing a new mode like bike-and-ride) are reflected in ActivitySim's demand model — not network coding in the traffic assignment model. He noted a comment he left two weeks ago may have been marked out of scope. Alex agreed this should be discussed further rather than resolved in the comments alone, and will flag it with Joe.
Amir also requested that any decision-making meetings on guide content be scheduled at Australia-friendly times. Alex committed to raising this with Joe.
6. Random Number Generation Testing
A report-out on RNG testing is scheduled for next week. Early indications from David suggest results are encouraging. This work is part of the broader EET effort led by RSG with support from Driftless andWSP, focused on replacing the legacy RNG components.
7. Admin — Phase 12 Agreements
Phase 12 (July 1 start) agreements are going out. The goal is to proceed with existing language and incorporate agency feedback in the 2027 cycle. Agencies with hard sticking points can raise them now.
Engineering meeting timing: Amir noted that engineering meetings are fixed at 4 AM Sydney time, unlike the executive meetings which rotate. He asked whether the engineering meeting could be shifted to alternate between Australia-friendly and US-friendly times. Alex committed to raising this with Joe.
Action Items
Notes prepared from transcript. Please flag any corrections.