[RBFOpt] [EXT] Re: RBFopt : question about noisy oracle definition

[NICOLAS Frederic]

Thank you Giacomo,

I am on implementing the call to the prediction of an RBF/GP model build with scikit-learn.
Indeed there is a parameter "return_std" in the "predict" function that, when set to True, forces to return a sigma value that can be used to define lower/upper variations (- or + 1.96 * sigma). I guess this is what is exactly required in the last two return positions of the obj_funct_noisy function.

Cordialement, Regards,

Frédéric NICOLAS
IFP Energies nouvelles
Direction Sciences et Technologies du Numérique / Département Modélisation Moteurs et Véhicules
Digital Sciences and Technologies Division / Engine and Vehicle Modeling Department
1 et 4, avenue de Bois-Préau
92852 RUEIL-MALMAISON
FRANCE
tel. : +33(0)14752-6696
email : frederic.nicolas at ifpen.fr


-----Message d'origine-----
De : Giacomo Nannicini [mailto:giacomo.n at gmail.com] 
Envoyé : vendredi 22 janvier 2021 16:50
À : NICOLAS Frederic
Cc : rbfopt at list.coin-or.org
Objet : [EXT] Re: [RBFOpt] RBFopt : question about noisy oracle definition

Hi Frederic,
both are viable options. The surrogate models built by RBFOpt are not
originally meant as predictors, therefore it is a bit clunky to build
them -- but it can be done. They also do not give estimates of the
error ranges. So I suspect it might be easier to use scikit-learn.
Regarding producing estimates of the error range (required as part of
the return values of the noisy objective function), if the estimator
doesn't provide a confidence interval, you can try to heuristically
set it to some arbitrary error range and adjust it experimentally.

Giacomo

P.S: GitHub discussions very recently became available for the
project, so you can use GitHub to ask questions now, if that's more
convenient.

On Fri, Jan 22, 2021 at 10:43 AM NICOLAS Frederic
<frederic.nicolas at ifpen.fr> wrote:
>
> Dear RBFOpt support team,
>
>
>
> I have been using RBFOpt for a few days and I have a question about the definition of the noisy oracle.
>
>
>
> In my case, I have defined my simulator (which is exact and slow to execute) as the one that is called through an implementation in the obj_funct that is given as a parameter of the RbfoptUserBlackBox function.
>
> On the other hand, I have a database of variants for which I already have the results (delivered earlier by my simulator) and I would like to build a surrogate model of these results to serve as the noisy oracle that I intend to implement as the obj_funct_noisy as the last parameter of the RbfoptUserBlackBox function.
>
>
>
> I was wandering whether I could build this surrogate model with some RBFOpt existing function (that returns the 3 awaited values) or if I should use an external library to this purpose, like scikit-learn for instance.
>
> Would you bring me a piece of advice on this ?
>
>
>
> Thank you for your kind support.
>
>
>
> Cordialement, Regards,
>
> Frédéric NICOLAS
> IFP Energies nouvelles
> Direction Sciences et Technologies du Numérique / Département Modélisation Moteurs et Véhicules
>
> Digital Sciences and Technologies Division / Engine and Vehicle Modeling Department
>
> 1 et 4, avenue de Bois-Préau
> 92852 RUEIL-MALMAISON
> FRANCE
> tel. : +33(0)14752-6696
> email : frederic.nicolas at ifpen.fr
>
>
>
> __________________________
>
> Avant d'imprimer, pensez à l'environnement ! Please consider the environment before printing !
> Ce message et toutes ses pièces jointes sont confidentiels et établis à l'intention exclusive de ses destinataires. Toute utilisation non conforme à sa destination, toute diffusion ou toute publication, totale ou partielle, est interdite, sauf autorisation expresse. IFP Energies nouvelles décline toute responsabilité au titre de ce message. This message and any attachments are confidential and intended solely for the addressees. Any unauthorised use or dissemination is prohibited. IFP Energies nouvelles should not be liable for this message.
> __________________________
>
> _______________________________________________
> RBFOpt mailing list
> RBFOpt at list.coin-or.org
> https://list.coin-or.org/mailman/listinfo/rbfopt
Courriel externe : Ce mail vous est adressé par un interlocuteur externe à IFPEN. Signalez ce message à l’assistance informatique assistance-informatique at ifpen.fr si l’adresse de l’émetteur ou le contenu vous paraissent suspects.
__________________________
Avant d'imprimer, pensez à l'environnement ! Please consider the environment before printing ! 
Ce message et toutes ses pièces jointes sont confidentiels et établis à l'intention exclusive de ses destinataires. Toute utilisation non conforme à sa destination, toute diffusion ou toute publication, totale ou partielle, est interdite, sauf autorisation expresse. IFP Energies nouvelles décline toute responsabilité au titre de ce message. This message and any attachments are confidential and intended solely for the addressees. Any unauthorised use or dissemination is prohibited. IFP Energies nouvelles should not be liable for this message.
__________________________