-
Notifications
You must be signed in to change notification settings - Fork 100
Project Meeting 2024.10.08
rcopperman edited this page Oct 8, 2024
·
1 revision
- General Admin - Phase 10A tasks
- Admin Transition
- Technical Tasks - Estimation Enhancements
- Phase 10A Tasks
- Consensus is to go after some of the tasks on list of issues
- Use Notes section to comment on list of issues
- Agencies should use Notes section to highlight super critical issues to address
- Admin Level of Effort
- MWCOG director is asking for level of effort document that he can review and use for further discussions with AMPO
- AMPO has reviewed document with Joe and Alex
- Mark made updates to table with meeting attendance, Joe and Alex will review
- Mark will share document as-is
- Mark will ask about timeline for further discussions with MPO directors and AMPO
- MWCOG director is asking for level of effort document that he can review and use for further discussions with AMPO
- Estimation Run Time Performance
- Estimation mode runs in single-process only
- Fix: Coalesced processes together ** Works if in estimation mode and multiprocessing is on
- EDB writing is slow
- Fix: Allow for EDBs to be written in csv, parquet, pickle file types
- New options:
- EDB_FILETYPE (options: csv, parquet, pkl)
- File type not available on all types of files
- New options:
- Fix: Allow for EDBs to be written in csv, parquet, pickle file types
- Destination Choice Location EDBs
- Fix: Allow for sampling
- SAMPLE_SIZE
- EDB_ALTS_FILE_FORMAT (options: compact, verbose)
- There does not need to be consistency between sample size in estimation and application
- Fix: Allow for sampling
- CI Testing
- Runs on small MTC test example
- Added infer.py into test system (calls internal ActivitySim functions for things like tour and trip IDs)
- Test all file types to ensure consistency
- Estimation mode runs in single-process only
- Larch Updates
- Enhance Larch useability and add data and output checking features
- New features check whether data input is in format that is useable for Larch
- Chosen alternative is formally not available (availability explicitly set to False)
- In a sized choice model, chosen alternative has zero size
- No choice made when case weight is non-zero
- NaNs appear in data
- Chooser attributes have insufficient variance
- Model is over-specified, with a nearly singular hessian (e.g. complete set of alternative specific constants)
- Upcoming: chosen alternative is functionally not available (e.g. alternative utility <-50 at null parameters)
- New features check whether data input is in format that is useable for Larch
- Enhance Larch useability and add data and output checking features
- Next Steps
- Finish implementation of location choice sampling
- Test Larch enhancements
- Benchmark improvements against pre-improvement platform with full-scale datasets
- Big Picture
- Task 1: Completed (or almost completed) 3 out of 4 main tasks
- Task 2: Working toward completion of 1 of 3 tasks
- Aiming to complete all tasks by mid-December