[Paradiseo-help] Hybridization of NSGA-II with Local search (Stochastic Hill Climber) procedure

Johann Dréo johann at dreo.fr
Ven 22 Mar 17:58:50 CET 2013


As the EO types are templatized, you could use the same EOT in both algorithms.

Then, I would design an evaluator wrapper that returns one of the
objectives, something like:
    moeoSplitEval::operator()( MOEO & indiv ) { return
indiv.objectiveVector()[1]; }

You should also add a scalar accessor to your own EOT:
    EOT::fitness( ScalarFitT& scalar) { EOT::ObjectiveVector ov; ov[0]
= scalar; ov[1] = whatever; this->fitness(ov); }

Then you use this evaluator to build you local search.

This would be nice to have a generic class to do that, do not hesitate
to send a patch if you manage to do it :-)

--
Johann

On Fri, Mar 22, 2013 at 4:56 PM, Nuno Leite <nleite at cc.isel.ipl.pt> wrote:
> Hi ParadisEO team,
>
> I'm a PhD student from Lisbon, Portugal, and I'm using your framework
> ParadisEO for multiobjective optimization.
>
> Hi have the following implementation problem:
>
> - I'm using framework's NSGA-II to solve a bi-objective optimization problem
> - Now, I want to hybridize NSGA-II with a local search metaheuristic,
> namely, Stochastic Local Search (SHC), in order to optimize some chosen
> solutions but considering just one of the objectives (single objective
> optimization).
>
> How can I do this?
>
> My problem is that the NSGA-II evaluation function considers the 2
> objectives but in the SHC I must consider only one objective. Can I do this
> just by parametrize the eval functions in the algorithms or should I create
> a Single Objective EO from the 2-Objective MOEO and optimize the transformed
> solutions with SHC and convert back in the end, inserting the improved
> (transformed from EO solutions) solutions again in the NSGA-II population?
>
> Thank you in advance.
>
> --
> Kind regards,
> Nuno Leite
>
>
> _______________________________________________
> Paradiseo-help mailing list
> Paradiseo-help at lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/paradiseo-help




Plus d'informations sur la liste de diffusion Paradiseo-help