Accéder au contenu.
Menu Sympa

starpu-devel - Re: [Starpu-devel] Problem with performance model calibration

Objet : Developers list for StarPU

Archives de la liste

Re: [Starpu-devel] Problem with performance model calibration


Chronologique Discussions 
  • From: Samuel Thibault <samuel.thibault@inria.fr>
  • To: Mirko Myllykoski <mirkom@cs.umu.se>
  • Cc: starpu-devel@lists.gforge.inria.fr
  • Subject: Re: [Starpu-devel] Problem with performance model calibration
  • Date: Wed, 10 Jun 2020 12:58:50 +0200
  • List-archive: <http://lists.gforge.inria.fr/pipermail/starpu-devel/>
  • List-id: "Developers list. For discussion of new features, code changes, etc." <starpu-devel.lists.gforge.inria.fr>
  • Organization: I am not organized

Mirko Myllykoski, le mer. 10 juin 2020 12:53:06 +0200, a ecrit:
> On 2020-06-09 21:24, Samuel Thibault wrote:
>
> > Could you try that with your application? I believe the result should
> > be much better.
>
> Thanks!
>
> It is much better now but still not perfect. Please see the attachment.

Indeed. The problem is that least square is not meant to work with a
constant, and linearization is not meant to include it either. The
dichotomy used in StarPU is a quick&dirty way to get it working
somewhat, but that's basically not theoretically grounded. Only real
proper multilinear regression with a dgels can catch this, so we'd have
to see why mlr doesn't work for your dataset.

Samuel




Archives gérées par MHonArc 2.6.19+.

Haut de le page