Add support for Riemanian objective through JuMP

[blegat]

Closes #273

[kellertuer]

I just finished #449 which does what you need (hopefully) at least for all objectives that (only) need cost/grad/hess – but it could in principle also be done for the non smooth prox ones.

So if you merge master into yours, you have even an -obj that creates the ScaledManifoldObjective wrapper :)

[kellertuer]

Thanks for continuing here!

Over easter I can take a look at the failing tests and try to maybe even write a more generic tutorial that uses this new mode.

Would it make sense to also take a look at the non-array-manifolds here already or would it be better to check that on a next PR?

[blegat]

I prefer smaller PRs whenever possible

[kellertuer]

Also perfect with me. Then we should fix the tests and maybe check with a small start of a tutorial, that all the new code works as expected.

[kellertuer]

The objective is an optimization part so it is part of Manopt not ManifoldsBase.
At the end we also need a Changelog entry for the new patch version (that is one of the tests failing).

I had hoped I can edit this locally and contribute a tutorial to this PR as well, that seems to not be the case I can only propose changes here (but maybe accept them to see how far CI gets?)

[blegat]

Weird, "Allow edits and access to secrets by maintainers" is checked

[kellertuer]

Sure I can pull and I can accept these edit comments here. But doing those feels a bit like hand-craftedly sending postcards to GitHub ;) I can not push from my locally checked out branch.

Whatever, then we do this PR shorter – you like that anyways :) - and just do the feature improvement. I can try locally already to get a tutorial working and that will be a next PR.

For not what is missing is

• Documenter is missing MathOptInterface.set which should go to the extension page docs/src/extensions.md in the right section
• I could not figure out the CI failure and conversion errors the new code still does.

[kellertuer]

Ah and the Changelog.md in the main folder needs a short comment for the new version that the interface was generalised.

[blegat]

What can we do if model.objective is nothing. Is there an easy way to get a zero objective to be used as second argument to decorate_objective! ?

[kellertuer]

What would that be used for? I currently do not have in mind what a “zero objective” mightt be.

The easiest “empty” objective is probably https://manoptjl.org/stable/plans/objective/#Manopt.ManifoldCostObjective so something like

obj = ManifoldCostObjective( (M,p) -> -Inf )

(or any other constant that -Inf)

[blegat]

It seems there is an issue with scaled objective because e98ea6f fixed the tests

[blegat]

With the scaled objective that you have added, the point does not move, very weird

[blegat]

The tests are passing locally, I won't have time to do more today but feel free to merge if ci is green in order to iterate on a follow up PR

[blegat]

The issue with the current approach is that the flipping of the objective sense don't happen if a riemannian objective is set

[kellertuer]

The tests are passing locally, I won't have time to do more today but feel free to merge if ci is green in order to iterate on a follow up PR

Nice! For me passing tests is fine enough, because we also have a test for the extension so it still works as expected. And I can try to write a short quarto-based tutorial and link to it from the extensions page as well in a next PR

The issue with the current approach is that the flipping of the objective sense don't happen if a riemannian objective is set

That I do not fully understand but I will check the code.

Thanks for all your work, would be super cool to maybe also get the other open point with non-array things done somewhen then – I can also first check how much I understand of the things that have to be changed then.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.88%. Comparing base (6430b31) to head (79b7bf9).
Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #448   +/-   ##
=======================================
  Coverage   99.88%   99.88%           
=======================================
  Files          83       83           
  Lines        8742     8747    +5     
=======================================
+ Hits         8732     8737    +5     
  Misses         10       10           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.