Réponse de Ricardo au sujet des méthodes de calibrar qu'il est possible de faire tourner avec des simu en parallel, ce qui n etait pas le cas de CMA-ES (question postée en 2020...).

Donc tjs pas CMA-ES mais apparemment pas mal d'autres. Je comprends qu'il faut notamment tester RvMMin (qui a en plus la fonction restart si ca s'arrete avant convergeance a cause des limites de temps de datarmor).

J'essaie bientot
-------- Message transféré --------
Sujet : Re: [roliveros-ramos/calibrar] CMAES does not run in parallel (#5)
Date : Wed, 14 Feb 2024 14:30:59 -0800
De : Ricardo Oliveros-Ramos <notifications@github.com>
Répondre à : roliveros-ramos/calibrar <reply+AN2EK64VFSBCGSET7V5DCDOD3JY2HEVBNHHB62XHG4@reply.github.com>
Pour : roliveros-ramos/calibrar <calibrar@noreply.github.com>
Copie à : slehuta <slehuta@ifremer.fr>, Author <author@noreply.github.com>


The last version of the package allows to read/write to the disk for all 19 methods currently implemented (so it always creates the appropiate i0, i1, ..., in folders needed to try different solutions). For clusters with schedulers like DATARMOR, the method 'Rvmmin' also has the restart functionality. This method uses derivatives, so can be used with deterministic models and in those cases should be way faster than AHR-ES, that was developed for stochastic models. We are not planning anymore to modify the code of CMA-ES so it can be run in parallel, but many other methods do provide the functionality and are included in the package.


Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: <roliveros-ramos/calibrar/issues/5/1944842020@github.com>