Unification with other Julia test databases

[anriseth]

There are multiple test databases floating around in the Julia community and it works be nice to reuse these amongst different packages.

For example, the JuliaSmoothOptimizers group has large collections of model problems that can be accessed with a unified interface using NLPModels

I propose that we look at a way to call such problems from Optim (either with an interface to optimize or by converting NLPModels to the Optim problem structure).

Most of the problems we have in this package are already in CUTEst.jl or other collections at JuliaSmoothOptimizers, so it may not be as useful to provide an API the other way (Optim->NLPModels).

[anriseth]

We could just rewrite all the problems in here to NLPModels.

Pro:

• Then they are more universally accessible

Con:

• This package would get a dependency that may mess up our unit tests in NLSolvers.

[pkofod]

It would certainly be nice not to have a full copy of different test problems in different packages, that's for sure.

[abelsiqueira]

cc @dpo @Xx-fractale-xX

Hello, we have three collections of problems:

• CUTEst.jl, which uses NLPModels;
• OptimizationProblems.jl, using JuMP, that we convert to NLPModels afterward;
• NLSProblems.jl, which uses JuMP, but need NLPModels, for Nonlinear Least Squares problems.

Currently, they don't pass on 1.0 because of the JuMP transition and FastClosures PR waiting to be merged.

NLPModels is being separated into a more essential part, using ForwardDiff, and the part using JuMP - NLPModelsJuMP, so supporting NLPModels would require NLPModels, LinearOperators, FastClosures and ForwardDiff. For testing using OptimizationProblems you then add JuMP and NLPModelsJuMP.

We would be very happy to see support for NLPModels conversion to Optim. It should be straightforward to access the required functions, but fell free to ask anything. If you'd like to try something more advanced, like using NLPModels internally, let me know.

Also, check our site: https://juliasmoothoptimizers.github.io/

[anriseth]

Thanks for your input, and the update on future plans :)