Local MINLP solver

[asavasci]

Hi, I am trying to solve an MINLP problem, but I am getting error output below

LoadError: Problem is a MINLP and no MINLP local solver given; use minlp_solver to specify a MINLP local solver in expression starting at...

I tried to use Juniper as local minlp_solver, however, it seems not working with JuMP v0.18, and if I upgrade JuMP, Alpine will not be working.

What would you suggest to work around this?

[ccoffrin]

This is indeed an issue with the incompatible versions of JuMP that are supported by Alpine and Juniper. You can try pinning Juniper to v0.3, which was the last version that supported JuMP v0.18, and it should should still with the current version of Alpine.

[asavasci]

I pinned JuMP and Juniper as follows [4076af6c] JuMP v0.18.6 ⚲
[2ddba703] Juniper v0.3.0 ⚲

I am still getting same error message

[ccoffrin]

According to this file, https://github.com/lanl-ansi/Alpine.jl/blob/master/Project.toml you might give Pavito.jl a try.

@Wikunia you might want to see if there is an easy way to use an old version of Juniper with Alpine.

[Wikunia]

@asavasci can you give an example of the part that is not working? I used release-0.3 of Juniper, Alpine v0.1.10, JuMP v0.18.6, Ipopt v0.6.1 and Cbc v0.6.6 and

# Here goes the building of your model...                                                                                                                          
setsolver(m, AlpineSolver(minlp_solver=JuniperSolver(IpoptSolver()), mip_solver=CbcSolver()))  

[asavasci]

After setting, setsolver(m, AlpineSolver(minlp_solver=JuniperSolver(IpoptSolver()), mip_solver=CbcSolver()))

Now, I am getting LoadError: type Variable has no field head in expression starting at solve(m) command

[harshangrjn]

@asavasci Can you share your exact problem you are solving (assuming the problem is a short one). I think I know the issue you are having.

[asavasci]

Here it is

test_alpine.txt

[harshangrjn]

@asavasci There were a few syntax and NLexpression issues which I fixed, after which the LoadError you had doesn't appear anymore. Also, global solvers typically perform better when you have variables initialized with finite bounds, rather than constraints. Here is the updated file: test_alpine_1.txt

However, Alpine has some issue with the lower bounding problem and isn't able to prove global optimality due to infeasibility (to do with numerics). But, Juniper is able to produce a feasible solution which is also global as it is easy to verify for your problem (and verified with another global solver). So, for now, feel free to run the attached file. Will keep you posted once the issue in Alpine is fixed.

[asavasci]

Noted. Thank you very much.

[odow]

This should be fixed now?

[harshangrjn]

Let me check on it and get back. It's been a while.

[harshangrjn]

Local solver isn't an issue in this anymore.