Re: weighted residuals are not real

[Ben Abbott]

On Nov 18, 2011, at 4:31 PM, Olaf Till wrote:

> On Fri, Nov 18, 2011 at 10:43:59AM -0500, Ben Abbott wrote:
>> 
>> On Nov 18, 2011, at 9:54 AM, Preeti Gaikwad wrote:
>> 
>>> now to solve this problem and to get best fit I want to remove some points 
>>> from x and y like from x i want to remove bad points
>>> 1.44 1.64 1.71 1.84 1.87 1.95 2.17
>>> 
>>> total 7 points and same the colum from y 
>>> then u see I am getting this error......thanks a lot for ur help and work
>> 
>> Ok. I see the error now. I'm not familiar with the leasqr.m code, but from 
>> what I infer it looks like a bug.
>> 
>>      http://octave.sourceforge.net/bugs.html
>> 
>> I've attached an archive with all the files needed  to reproduce the error 
>> with run.
>> 
>> Ben
> 
> Thanks for preparing this, but why do you think it is a bug? The error 
> happens when
> leasqr tries the parameters [3.79885888315165; 2.53178132967943], which are 
> inside
> the bounds. The users model function returns non-real residuals for these 
> parameters,
> and this is reported with an error. I can't see a bug. Am I missing something?
> 
> Olaf
> 

When I ran the example and saved the "pin" (passed to __lm_svd__.m) that 
resulted in non-real residues, pin = [0.1 3.0] which produce should produce 
real residues.

I didn't bother to dig into the details of how leasqr works, so I may have 
overlooked something.

Ben