[SciPy-user] Eror estimation in leastsq

Robert Kern kern at caltech.edu
Wed Feb 20 15:25:49 CST 2002


On Wed, Feb 20, 2002 at 10:11:10AM -0500, Travis Oliphant wrote:

[snip]

> They solve two different problems.  The one you mentioned solves the
> linear least-squares problem, while the one in minpack is much more
> general.  I don't remember exactly how to get residuals from the minpack
> algorithm.

I don't think that solves the original problem, though. Nadav requested
the estimated standard deviations of the estimated parameters, not the
residuals of the data to the fit curve. I don't think you can go from
one to the other (IIRC).

> We still need to add a "total least squares" algorithm with its
> accompanying error estimates.

http://www.netlib.org/odrpack is probably the way to go. My wrappers
work, for the most part, but I still need to track down a segfault when
running multiple fits.

It's hand-written C loosely modelled after the minpack wrappers. At the
time, I didn't think that f2py had the capabilities I needed. I should
rethink that now.

Should some free time magically appear, I will hunt down the remaining
bugs and submit the module to you all for inclusion in SciPy. If anyone
else is interested, mostly up-to-date source is at
http://starship.python.net/crew/kernr/source/odr-0.5.1.tar.gz

> -Travis

-- 
Robert Kern
Ruddock House President
kern at caltech.edu

"In the fields of hell where the grass grows high
 Are the graves of dreams allowed to die."
  -- Richard Harter



More information about the SciPy-user mailing list