elainethale
commented
5 days ago @Nongchao, I did my best in minimal time to merge your latest changes into mine. Because of my surface-level understanding you'll probably find the current code unsatisfying and in need of more clean up for the reasons you outline above. In short:
- I am currently observing lower-level problem failures during the first try, but successes after that (e.g., fail for iteration 2026-1, but success for iteration 2026-2). That is, the current code runs for the whole simulation horizon, but could be more efficient.
- I didn't completely roll back my changes because:
- It's more flexible to key off of the individual agents as to which was their last solve year, rather than assuming we'll always run agents in their current order
- I was also uncomfortable with changing model_data.index_y in multiple places (i.e., once in the main iteration loop and twice in ipp_group), and indeed that is why there are deepcopy(model_data.index_y)'s all over the place in ipp_group (which could maybe be removed now???)
To get toward a full draft analysis, I'm going to merge to main now, but we should discuss next week and probably do one more iteration to get the lower level solving cleanly again.
@Nongchao, my last commit seems to be successful -- 2023 WM, Flat is running smoothly now. Take a look/try running and approve if you can confirm? I'm also running all 15 years on Kestrel ...