Thread Subject: lsqcurvefit versus fminsearchbnd

When is it appropriate to use lsqcurvefit versus fminsearchbnd?

I am fitting around 11 parameters, and lsqcurvefit seems to be superior. I am simply inquiring why.

Thanks

On 11/3/2011 3:39 PM, Matthew wrote:
> When is it appropriate to use lsqcurvefit versus fminsearchbnd?
>
> I am fitting around 11 parameters, and lsqcurvefit seems to be superior.
> I am simply inquiring why.
> Thanks

The algorithm in lsqcurvefit uses gradient information, and is usually
much faster and more reliable. fminsearchbnd is based on fmransearch,
which uses the Nelder-Mead simplex algorithm. This algorithm does not
use gradients, is poor in more than a few dimensions, and is not
guaranteed to converge.

Alan Weiss
MATLAB mathematical toolbox documentation

"Matthew" <matthew.merritt.nospam@utsouthwestern.edu> wrote in message <j8uqkv$6rg$1@newscl01ah.mathworks.com>...
> When is it appropriate to use lsqcurvefit versus fminsearchbnd?
>
> I am fitting around 11 parameters, and lsqcurvefit seems to be superior. I am simply inquiring why.
>
> Thanks

In addition to Alan's comments, I will NEVER use fminsearch, or its
bounded cousin, fminsearchbnd, on large problems. "Large" is of
course in the eye of the beholder, but a typical upper limit of
practicality is somewhere around 6 variables, with 3 variables a
better realistic limit for efficient behavior.

11 variables is well beyond my limit in any case.

John