Re: polyfit() and Inf

[Paul Kienzle]

stefan wrote:

> I was experimenting with the polyfit() function.  It works correct with
> "normal" functions (y = f(x)).  But it seems to compute ages (I never
> waited until it finished) when there is a 'Inf' in the dataset.
>
>         x = linspace(0,1,100);
>         y = x * 1 + 4;
>         y(99) = Inf;
>         [p,r] = polyfit(x,y,1);
>         disp(p);
>
> This piece of code works fine without the 'y(99) = Inf;' line and stucks
> with it.
>
> I implemented the polyfit() function on my own.  I get infinite results
> but it does not stuck!  Is this a real problem/bug with polyfit()?  I
> could send you my implemetation, if interested...

I get the same result (cygwin, octave-2.1.42).  This is
probably a bug in \, which relies on the functions in
LAPACK.

wpolyfit  in octave-forge returns  NaN:

http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/~checkout~/octave/octave-forge/main/optim/wpolyfit.m?rev=HEAD&content-type=text/plain

It uses the slower but more accurate QR decomposition:

 [Q,R,P] = qr(X,0);
 p = R\(Q'*y);
 p(P) = p;

You can check it yourself against the NIST
Statistical Reference Datasets

http://www.itl.nist.gov/div898/strd/

Paul Kienzle
address@hidden




-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------