Use JuMP.fix_discrete_variables to calculate duals (fallback)

[manuelma]

JuMP provides this function since recently and it would be good to at least try it out instead of our rather sophisticated dual solution in another thread...

[DillonJ]

Good idea. As an aside, I see JuMP also now supports setting some solver options generally, such as set_time_limit_sec and maybe there are others. We could support these as generic model-level parameters instead of / in addition to the solver-specic parameters using the mip_solver_options and lp_solver_options map parameters

[manuelma]

Good idea @DillonJ - although about using JuMP.fix_discrete_variables... looking at the code it doesn't seem too impressive. They do the obvious thing we are already doing with the copy of the model, but a little of extra work to set lower and upper bounds for binary variables that don't have a value for some reason (which is never our case). So I'm not convinced.

At the same time, our approach generates a lot of copies of the model - but they are eventually cleaned up... I thing we need to carefully evaluate pros and cons...

[DillonJ]

Might need to test the performance on a real-sized problem