feat: time limits and solution statuses #89
Conversation
|
@mmghannam you might be interested in this |
|
cool ! |
done ! |
|
@KnorpelSenf , you have been pushing a lot of great PRs lately. Would you like to be granted commit access to this repo, in order to participate to its maintenance on the long run ? |
Sure thing, thank you! |
KnorpelSenf
changed the title
| let sol = pb.solve().unwrap(); | ||
| assert_float_eq!(sol.value(v), limit, abs <= 1e-8); | ||
| } | ||
|
|
||
| const VARIABLE_COUNT: usize = 75; | ||
| const CLAUSES: &[(i32, i32, i32)] = &[ |
There was a problem hiding this comment.
Please let me know if you have a better idea than this
|
@lovasoa please check out rust-or/highs#24. After it is released, I can revert 66662fb and then this is ready. |
| TimeLimit, | ||
| /// The solution is not optimal and it was obtained because the gap limit | ||
| /// was reached | ||
| GapLimit, |
There was a problem hiding this comment.
I would ideally like to do
| GapLimit, | |
| GapLimit(f64), |
to communicate the actual gap of the solution but I think the underlying bindings for CBC and SCIP don't expose this data yet.
I mostly think this would be better library design but I honestly don't ever need this myself … so I might just not do it?
|
Marking this as draft to make sure this isn't merged until we release highs |
|
Feel free to undo when the dependencies point to stable highs |
Probably wise, I'll do that myself from now on |
It is currently possible to set time limits for individual solvers.
This pull request introduced a generic API to set this for all solvers that provide implementation for this. Currently, CBC, HiGHS, and SCIP are supported. HiGHS bindings depend on rust-or/highs#24, so this PR temporary switches to HiGHS from GitHub. As soon as rust-or/highs#24 is merged, this commit can be reverted.
These changes also include a way to read the solutions even when the solving process was halted early due to a time limit. A new status on the solution lets developers differentiate between optimal solution and solutions reached via time limit.
Also, it aligns how MIP gaps are handled with the newly introduced solution status.