search

jump-dev / JuMP.jl

Modeling language for Mathematical Optimization (linear, mixed-integer, conic, semidefinite, nonlinear)
http://jump.dev/JuMP.jl/
Other
2.19k stars 390 forks source link

Don't ask for reduced costs until user does? #483

Closed tkelman closed 9 years ago

tkelman commented 9 years ago

Running OsilBonminSolver against JuMP master's test/nonlinear.jl results in an undefined reference during the first MINLP test, since Bonmin doesn't give duals (at least not in the results osrl file) for MINLP's. The test doesn't actually ask for the duals, so seems like the error could be avoidable.

(p.s. this also means that I hope you don't tag before my talk, or my test example might not work)

mlubin commented 9 years ago

We probably shouldn't ask for duals at all when solving MINLP. Don't worry, not tagging soon. On Jul 15, 2015 1:53 AM, "Tony Kelman" notifications@github.com wrote:

Running OsilBonminSolver against JuMP master's test/nonlinear.jl results in an undefined reference during the first MINLP test, since Bonmin doesn't give duals (at least not in the results osrl file) for MINLP's. The test doesn't actually ask for the duals, so seems like the error could be avoidable.

(p.s. this also means that I hope you don't tag before my talk, or my test example might not work)

— Reply to this email directly or view it on GitHub https://github.com/JuliaOpt/JuMP.jl/issues/483.

tkelman commented 9 years ago

That should be okay for now, but there might be continuous solvers that don't always give duals. If the user doesn't ask for them should there still be an error?

mlubin commented 9 years ago

That's tricky to do with the current infrastructure. I think the cleanest approach right now would be for the solver to return a vector of NaNs if duals aren't available. On Jul 15, 2015 3:15 PM, "Tony Kelman" notifications@github.com wrote:

That should be okay for now, but there might be continuous solvers that don't always give duals. If the user doesn't ask for them should there still be an error?

— Reply to this email directly or view it on GitHub https://github.com/JuliaOpt/JuMP.jl/issues/483#issuecomment-121718630.