On Fri, 20 Jul 2007, John Gehman wrote:

[usefulness of sigma]

> For the case where all data has the same error, as yours appears to,  
> then all data is weighted equally. HOWEVER, this value does scale the  
> residual and the sensitivity of the gradients, and I believe  
> therefore does influence convergence. It may not make that big a  
> difference in the regime where you are working, but I have noticed  
> before with some of my applications that it was important to be  
> realistic about the error.

OK, I thoughts so. Effectively, it allows the fitter to get it slightly 
wrong before cranking up the error.

Is there a way to supply a custom error function (e.g. via a callback) to 
the fitter in GSL?

> I'm not sure whether a gsl_vector_scale after the fact  
> would be more efficient that explicit divisions during the loop?

I'm not sure, but I think that would produce a different result...

Gordan



_______________________________________________
Help-gsl mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/help-gsl

Reply via email to