Re: Scilab is now GPL v2 compatible - as the site claims

[Sergei Steshenko]

--- On Sun, 9/21/08, Marc Normandin <address@hidden> wrote:

> From: Marc Normandin <address@hidden>
> Subject: Re: Scilab is now GPL v2 compatible - as the site claims
> To: "address@hidden" <address@hidden>
> Cc: "address@hidden" <address@hidden>
> Date: Sunday, September 21, 2008, 12:19 AM
> Sergei Steshenko wrote:
> > Anyway, frep2tf has a subtle, though not serious,
> flaw.
> > 
> > IIRC, it uses sum of abs deviations rather than sum of
> squares of abs
> > deviations as its error metric. I mean, strictly
> saying, it's not least
> > squares approximation as it claims.
> 
> From your description, it sounds like the the method is
> reasonable,
> though perhaps not what people would expect; thus, it's
> the
> documentation that's incorrect.  They've simply
> chosen a "robust"
> estimator in favor of the the minimum mean square approach,
> which one
> can argue is sensitive to noise.
> 
> Regards,
> Marc
> 
> -- 

Again, I do not understand all the details, but, I think, they are
solving a system of linear equations based on true least squares approach
(this happens in each iteration), but they choose the best solution after
a number of iteration based on the sum of abs deviations, not on the sum
of squares of abs deviations.

So, I think, their approach has an internal contradiction.

Regards,
  Sergei.