How to get the Solution Pool using Julia/CPLEX?

[GiAbreu]

Hi everyone,

I need the solution pool for storing multiple solutions of a MIP model. I using Julia and the packages JuMP and CPLEX as a solver. So I need to collect solutions within a given percentage of the optimal solution but I don’t know how to do it in Julia, anybody can help me, please?

[IssamT]

I don't think it's available on the JuMP layer since it's solver specific. If you are ready to do a bit of hacking, then you can maybe try using setincumbentcallbackfunc of CPLEX.jl

[GiAbreu]

I set the parameters as explained on the IBM website, but I don't know how to call the populate solution pool. Using callbacks I only got the values ​of the objective function, best bound, numbers of nodes of B&B tree, but I need the values ​​of the multiple solutions for the variables that have the same objective function value, which means populate the solution pool.

[odow]

We now wrap CPXpopulate(model.env, model.lp), but there is still no nice way of accessing multiple solutions via MOI. Ideally, we would have something like the following:

model = CPLEX.Optimizer()
x = MOI.add_variable(model)
# ... model definition ...
MOI.optimize!(model)
CPXpopulate(model.env, model.lp)
for n = 1:MOI.get(model, MOI.ResultCount())
    x_n = MOI.get(model, MOI.VariablePrimal(n), x)
    println("Value of x in solution $(n) is $(x_n)")
end

This is all possible via the exposed C API, so it just needs some plumbing. I won't be implementing this, but I will leave the issue open for a motivated person in the community.

[davide-f]

Hi, are there updates on this issue? More and more solvers are now providing solution pools and it would be an useful feature
Thank you

[odow]

@jd-foster worked out the details here: jump-dev/jump-dev.github.io#58. It just needs implementing at the MOI level. PRs accepted.

[odow]

Closing this as won't-fix. It has been a few years and there doesn't seem to be much demand for this. It also requires some explicit C API calls, rather than the option setting like Gurobi which makes it a bit more annoying to implement.

@jd-foster has written a tutorial that demonstrates how to get multiple solutions with CPLEX:
See https://jump.dev/tutorials/2021/11/02/tutorial-multi-jdf/

If you are wanting multiple solutions with CPLEX and that isn't sufficient, please comment below and we can consider re-opening this issue.

[chriscoey]

@odow this was closed a few months back, so I imagine the story hasn't changed since then. Do you have a sense for how much work it would be to expose this in a more user-friendly way here? Even better would be if we could have better support in MOI for getting multiple solutions from MIP solvers.

[odow]

The issue is that CPLEX doesn't neatly provide multiple solutions. You need to create an populate a separate solution pool object, etc.

MOI has great support for querying multiple solutions. See, e.g, Gurobi. The complexity is on the solver side, not JuMP or MOI.

I would consider PRs to CPLEX.jl, but this is not a priority for me.

[chriscoey]

Thanks, makes sense. I guess by suggesting something at the MOI level I meant more around the concept of solution pools, but you are saying that just the concept of multiple solutions is enough for MOI, as evidenced by Gurobi.jl.

[odow]

as evidenced by Gurobi.jl

And also MiniZinc.jl, MultiObjectiveAlgorithms.jl, etc.

I don't want to add a solution pool concept to MOI. The solution here is to add some sort of special option to CPLEX.jl that, instead of solving normally, solves via the solution pool and makes the set of solutions available. It's perfectly do-able, but it requires some amount of work.

[odow]

Re-opening as maybe-planned