12
$\begingroup$

LocalSolver is a company which provides a global optimization solver, combining exact and heuristic techniques.

The benchmarks on their website are quite impressive. For example, they claim they can solve large TSPs within 10 seconds with gaps smaller than 5% in average :

enter image description here

  1. Does anyone have any experience with LocalSolver, and what are the pro's and con's ?

  2. Are there any peer reviewed publications where these benchmarks can be found ? It would be interesting to see if there are any comparisons with heuristics and metaheuristics. Localsolver performs local search after all !

$\endgroup$
5
  • $\begingroup$ I had a little search on their website and google scholar and found lots of papers talking about their approach in designing the solver(authors are the designers of the LocalSolvers) and found many application papers both in English and French(!). If you search for the designer's name (Thierry Benoist, Frédéric Gardi - these guys have google scholar accounts) you will find the source of all those papers. $\endgroup$ May 11, 2020 at 23:19
  • 2
    $\begingroup$ swmath.org/software/4850 <--- here is a list of academic papers about the solver. $\endgroup$ May 11, 2020 at 23:43
  • $\begingroup$ I also read somewhere that they (Localsolver) claim to have best solution approach to solve non-linear optimization problems in particular. $\endgroup$ May 12, 2020 at 6:16
  • 1
    $\begingroup$ Note also that LocalSolver is a user on this site. $\endgroup$ May 12, 2020 at 14:34
  • $\begingroup$ Yes, I think it is great if they can contribute somehow on this site. It would be interesting to compare their pros and cons, with users. Perhaps it could help them as well identify potential improvements. $\endgroup$
    – Kuifje
    May 12, 2020 at 14:41

1 Answer 1

13
$\begingroup$

Here is a quick summary of the pros and cons of Hexaly (ex LocalSolver), a global optimization solver combining exact and heuristic techniques. Please note that this summary is written by the Hexaly team, as asked by Kuifje in the comment above.

Pros:

  • You will model your problem using nonlinear and set-based mathematical modeling APIs. Then, your math model will be closer to the real-life problem, more concise, and easier to maintain. For an example of what means "set-based modeling" have a look at the TSP model described here. Hexaly also supports external functions in your model (also known as black-box optimization) and offers multiobjective modeling features.
  • Having modeled your problem for Hexaly, you will get solutions faster, and your model will be more scalable than MILP solvers, thanks to innovative heuristics. In addition, Hexaly offers (global) dual bounds.

Cons:

  • You must reformulate your problem by using the appropriate Hexaly modeling constructs and following Hexaly modeling best practices.
  • Don't give MILP-like models to Hexaly, especially LP or MPS files, in hopes of getting good results.

State-of-the-art MIP solvers remain better than Hexaly when solving pure linear problems and then nearly linear problems (that is, problems that allow good linear approximations). Because the LP codes (simplex, interior-point) used inside these MIP solvers are still stronger than those inside Hexaly (note that we work hard to decrease this gap each year). This performance gap is noticeable for large-scale, ill-posed linear programs. Now, when problems become more combinatorial (in particular: routing, scheduling, packing, assignment) or more nonlinear/nonconvex/nonsmooth in the continuous space, Hexaly is a tool of choice that deserves to be considered for your optimization project.

Additional non-technical strengths of Hexaly that our users mention: the dedicated and reactive support offered by the R&D team to model and solve at best your problems using Hexaly; an aggressive roadmap with two new releases per year with constant performance improvements and new features; a simple, competitive, flexible licensing & pricing.

For examples of problems modeled using Hexaly, have a look at https://www.hexaly.com/docs/last/exampletour/index.html.

Check out this link for some benchmarks against MIP solvers. You can reproduce the results in these benchmarks by registering and asking for free trial licenses on the Hexaly website.

Note that the technical papers related to Hexaly which you can find on the web are largely outdated. We don't publish anymore about Hexaly for several years (mostly for confidentiality reasons).

$\endgroup$
8
  • 1
    $\begingroup$ Does LocalSolver offer dual bounds, as well nowadays? That is, can it provide gap information? $\endgroup$ May 14, 2020 at 6:47
  • 1
    $\begingroup$ @LocalSolver Thanks for your feedback ! LocalSolver seems like a nice tool with many strengths. Also, I enjoyed your recent posts regarding the largest empty rectangle, nice stuff. $\endgroup$
    – Kuifje
    May 14, 2020 at 7:06
  • 1
    $\begingroup$ @RobertSchwarz Yes, LocalSolver offers (global) dual bounds as well. For example, have a look to localsolver.com/news.html?id=96. $\endgroup$
    – Hexaly
    May 14, 2020 at 9:52
  • $\begingroup$ @Kuifje Thanks a lot for your kind words and encouragements. This post was about recreative, funny problems but shows quite well LocalSolver capabilities as global optimization solver. $\endgroup$
    – Hexaly
    May 14, 2020 at 9:58
  • $\begingroup$ @LocalSolver: can you repost the link with the dual-bounds example, it doesn't work and I can't seem to see a reference to dual bounds in your examples or documentation. Are the dual bounds guaranteed to be accurate in all cases? $\endgroup$
    – ldog
    Nov 22, 2022 at 22:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.