Re: polyfit: add the ability to specify the polynomial template

[CdeMills]

John W. Eaton wrote
> 
> 
> I don't see it as a regression in any case, since the behavior has
> been the same for a long time now.
> 
> This change can maybe go on the default branch, but first think about
> whether it introduces potential compatibility problems with Matlab's
> polyfit function.  If so, then we should think carefully about whether
> we want the change as default behavior.
> 
> jwe
> 
If you use it 'the old way', you get results 'the old way'. I was very
carefull at not changing the way computations were done. The only
behavioural difference is that if you require the s matrix, you get a
supplemental field, s.C. BTW, as the qr may permute columns and the ordering
vector is not returned, getting s.R  doesn't help, as you don't get the
mapping between its inverse (the unscaled covariance matrix) and the
corresponding columns of the Vandermonde matrix, which are each linked to
one coefficient of the polynomial.

In my case, the new mask permit to choose between offset-free and models
with offset, and it makes differences.

I would also emphasise the fact that I switched to Octave (what was the
first version I used ? 1.something ? Is it possible ?) because I wanted to
have access to the sources, and be able to correct bugs and inaccuracies,
and see them integrated on the long term into the mainsteam code.

So if you prefer to see the patch applied to default and not stable, it's OK
if it percolates into stable on the midterm. 

Regards

Pascal


--
View this message in context: 
http://octave.1599824.n4.nabble.com/polyfit-add-the-ability-to-specify-the-polynomial-template-tp4357144p4365566.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.