Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fixing issue with post solve #191

Merged
merged 20 commits into from
Sep 5, 2024
Merged

Fixing issue with post solve #191

merged 20 commits into from
Sep 5, 2024

Conversation

dhendryc
Copy link
Collaborator

@dhendryc dhendryc commented Jul 2, 2024

In some instances, the post solve produced a worse solution than the current incumbent.

This was due to nearly integer points (e.g. 1 + 2e-10) being evaluated by a strongly convex (or sharp) objective function.

@dhendryc
Copy link
Collaborator Author

dhendryc commented Jul 2, 2024

  • Wait for benchmarks to pass

@dhendryc
Copy link
Collaborator Author

I added a function that is basically like a mini-postsolve for each new solution. This is, however, rather costly and thus disabled by default. This only really makes sense if we use a MIP solver, I think. The combinatorial BLMOs usually return exact integer vertices anyway.

@matbesancon I am not sure if this is the cleanest way to do it though.

@dhendryc dhendryc requested a review from matbesancon August 28, 2024 08:56
dhendryc and others added 3 commits August 28, 2024 15:22
BCG returns non integer vertices in FrankWolfe callback due to simplex descent steps.
@dhendryc dhendryc linked an issue Aug 28, 2024 that may be closed by this pull request
@matbesancon
Copy link
Member

what's the reason a simplex descent step would use a non-vertex? It should be using the same vertices as other lazy steps

@dhendryc
Copy link
Collaborator Author

dhendryc commented Sep 2, 2024

The function starting here https://github.com/ZIB-IOL/FrankWolfe.jl/blob/a53609b23768eae00b2c029c075acc01e8f5f6ee/src/blended_cg.jl#L896 does not necessarily return a vertex. From what I understand y is some point in the convex combination.

@dhendryc
Copy link
Collaborator Author

dhendryc commented Sep 4, 2024

Round the solutions of the subproblems.

@matbesancon matbesancon merged commit 4adbee4 into main Sep 5, 2024
3 checks passed
@matbesancon matbesancon deleted the postsolve-fix branch September 5, 2024 13:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

BCG non integer vertices
2 participants