SciML
Arbitrary constraints
Correct me if I am wrong, but it is not very clear in the documentation how would one go about defining constraints that are not "box" constraints. One of the keywords of OptimizationFunction is cons but it is not explained what is the syntax to define it. Is this feature implemented and the documentation is missing or the feature is not implemented yet?
Also, as of now, only ForwardDiff.jl can be use for constraints, is that correct?
Is this feature implemented and the documentation is missing
~That's the case, you can refer to the tests for now if you have a use case.~
It is documented here https://galacticoptim.sciml.ai/stable/optimization_packages/optim/#Local-Constraint, examples with MOI are not included in the docs yet but it should work just like Optim wrapper.
Also, as of now, only ForwardDiff.jl can be use for constraints, is that correct?
Yes that's correct.
Thank you, I had not seen this example.
So, is I understand well, the keyword cons is the constraint function and lcons and ucons are the value of the bounds. I presume that if one wants an equality constraint, one sets lcons==ucons. Is this correct?
No, a lot of optimizers won't like that. We need to add a thing for equality constraints and document this appropriately. It's just a work in progress.
@ChrisRackauckas has the new version of GalacticOptim released today implemented equality constraints? Still not sure how to implement them besides setting lcons==ucons