Re: scaling for polyfit.m [patch]

[Paul Kienzle]

On Aug 22, 2006, at 4:37 PM, John W. Eaton wrote:

> On  9-Aug-2006, Thomas Weber wrote:
>
> | some time ago we had a bug[1] filed in Debian against polyfit.m; it
> | breaks down for large values of x (which is known and a problem in
> | Matlab as well).
> |
> | Based on the idea of the bug reporter, I've changed the polyfit 
> function
> | to always apply scaling, making it work for large x values.
> |
> | The patch is attached; it includes a test with the data provided in 
> the
> | bugreport.
> |
> | One thing where I was unsure is which polynomial to use for 
> calculating
> | the values of S; I've taken the scaled polynomial because I think 
> most
> | people want the conditition value for the actual calculation.
>
>
> I'm not sure that we should always do the scaling.  Maybe it should be
> optional?

Given the general policy of accuracy before speed, is there any reason 
not
to do scaling?

- Paul