Re: Polyfit with scaling

[Thomas Weber]

Am Sonntag, den 03.02.2008, 19:20 -0500 schrieb Ben Abbott:

> Give these results and the need to remain compatible with past
> expectations, and with Matlab, I'm inclined to make both QR and
> normalization optional.

Why do we need to stay compatible with past expectations? If polyfit
gets better with fitting, I don't think anybody would mind.

With regards to Matlab, it uses a QR decomposition 
http://www.mathworks.com/access/helpdesk/help/techdoc/ref/polyfit.html


Leaving things as they are is in my opinion cleary what we should *not*
do. There have been complaints about polyfit going back to 1999
http://velveeta.che.wisc.edu/octave/lists/archive//help-octave.1999/msg01007.html


And frankly, pointing people to wpolyfit looks bad: "Oh, yes, polyfit
doesn't handle your data, but there's another function out there. We
know about it for ages, but the discussion lead nowhere and so we just
stay with the worst implementation." If wpolyfit is better (or if it's
just the QR decomposition), let's just agree on using that one. I'm
willing to change polyfit to QR. 

If we can get away without normalization, the better.

        Thomas