Support additional solvers

[staadecker]

Update, we now support COPT and Mosek

Currently, Pyoframe supports Gurobi, HiGHS, and Ipopt.

I will happily consider adding support for the following solvers if there is interest. React to this message to show interest.

• 🎉 Mosek solver
• 🚀 KNITRO solver
• COPT solver

Steps to add a new solver

If the solver is supported by PyOptInterface the process is relatively straightforward:

1. You'll need a license for testing and I (Martin) will need a license for testing and maintain the library + running the tests in the CI pipeline.
2. Add a _Solver object in the _constants.py file and make other changes to the source code such that all the tests succeed. This includes a) updating the model initializing code in _model.py, and b) generating new test output files so that test_examples.py passes. If the solver has quirks or behaves differently than other solvers, add additional feature flags to the _Solver class rather than directly adding edge cases to the core code.
3. Update the installation documentation page. (Run mkdocs serve to preview the documentation)
4. Add the solver and its license to the Github Actions CI pipeline.

[eduardosalaz]

Martin, quick question, COPT support hasn't been added yet right?

I'll work with the COPT developers from now on so I think I can find time to integrate the solver into Pyoframe, if it hasn't been integrated yet.

[staadecker]

Hi Eduardo! That's correct. COPT support hasn't been added so far and your help would be awesome! If you get started on the implementation I can look into getting a COPT academic license for the GitHub Actions CI tests.

[eduardosalaz]

Will start work in the coming days on COPT then, will let them know about the license for CI tests.

[eduardosalaz]

#181 pr in progress

[eduardosalaz]

KNITRO support was added recently. It's a more robust nonlinear solver than IPOPT. I think it's worth looking into it.

[eduardosalaz]

I can request a trial license for 1 month full-sized or limited size for 3 months. I think it's better if we wait for a bit for PyOptInterface to stabilize the new interface and I can start work on this.

[staadecker]

Awesome! Sounds great to me.

[Andreas-naut]

Mosek would indeed be nice, for giant sparse problems (and that's the reason I found pyoframe :-)) it's interior point is very competitive. Also it's really cheap compared to the others.