I would like to solve a non linear optimisation problem which is non smooth. I programmed the problem with JuMP and I tried to solve it with Ipopt solver. However, Ipopt is not able to solve it. I had a look about global solver and metaheursitic. I see the PSO (particle swarm optimization), and I searched if there is one with julia and JuMP interface, however I haven’t found any. I saw blackboxoptim.jl, but without PSO and JuMP interface.
Does someone has any information about global optimizer (PSO) and JuMP ?
Before jumping to PSO have you tried giving a global optimization solver a try? For example, Couenne, Baron, Alpine? For small to medium sized problems these can be quite effective.
Thanks M. Coffrin for answering,
I try to follow your guidance:
First, I had a look with Alpine, however the non-linear problem has exp(x) and log(x), it is written on github Alpine page that it is not supported. Second, I do not have Baron license nor my lab, and I am not working with the affiliate universities.
Finally, I tried Couenne with AmplNLWriter.jl. Unfortunately, after couple of hours of computation, there are still no outputs (more than 1 day). To compare, Ipopt return “Exit: Converged to a point of local infeasibility. Problem may be infeasible” after 24 minutes computing.
I had a look at optim.jl, there is a Particle Swarm Algorithm. Is there a way to translate JuMP model to be able to be optimize through Optim.jl?