Nonlinear Least Squares Fitting

Nonlinear Least Squares Fitting

[Andreas Gros]

Hello,

I'd like to submit a "bug" in the solver:
gsl_multifit_fdfsolver_lmsder.

In the section "Providing the function to be minimized" where the
function gsl_multifit_function_fdf is described, you mention the
function f that should store the distance between the values
calculated by the algorithm and the provided values. 
As I tested the algorithm I've found, that this distance is not taken
as a "square", instead the algebraic sign of the distance seems to be
of some importance. My results were quite different, whenever I took
(y[i]-Yi) instead of (Yi-y[i]) (as it is shown in the example).

Taking the latter the solver works fine, taking the first distance the
results have been quite funny.

Andreas Gros 


Andreas Gros, Tel.: 0179-1062416
Lotter Str. 99, 49078 Osnabrück
Amalienstr. 8, 97072 Würzburg
http://www.gros-webdesign.de

Re: Nonlinear Least Squares Fitting

[Brian Gough]

Andreas Gros writes:
 >  In the section "Providing the function to be minimized" where the
 > function gsl_multifit_function_fdf is described, you mention the
 > function f that should store the distance between the values
 > calculated by the algorithm and the provided values.  As I tested
 > the algorithm I've found, that this distance is not taken as a
 > "square", instead the algebraic sign of the distance seems to be of
 > some importance. My results were quite different, whenever I took
 > (y[i]-Yi) instead of (Yi-y[i]) (as it is shown in the example).
 >  Taking the latter the solver works fine, taking the first distance
 > the results have been quite funny.

Hi,
The sign of the jacobian needs to match the sign of the function.
If you change one you need to change the other.
I think this will fix your problem.
regards
Brian

Re: Nonlinear Least Squares Fitting

[Brian Gough]

Andreas Gros writes:
 >  In the section "Providing the function to be minimized" where the
 > function gsl_multifit_function_fdf is described, you mention the
 > function f that should store the distance between the values
 > calculated by the algorithm and the provided values.  As I tested
 > the algorithm I've found, that this distance is not taken as a
 > "square", instead the algebraic sign of the distance seems to be of
 > some importance. My results were quite different, whenever I took
 > (y[i]-Yi) instead of (Yi-y[i]) (as it is shown in the example).
 >  Taking the latter the solver works fine, taking the first distance
 > the results have been quite funny.

Hi,
The sign of the jacobian needs to match the sign of the function.
If you change one you need to change the other.
I think this will fix your problem.
regards
Brian

Nonlinear Least Squares Fitting

[Andreas Gros]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 851 bytes --]

Hello,

I'd like to submit a "bug" in the solver:
gsl_multifit_fdfsolver_lmsder.

In the section "Providing the function to be minimized" where the
function gsl_multifit_function_fdf is described, you mention the
function f that should store the distance between the values
calculated by the algorithm and the provided values. 
As I tested the algorithm I've found, that this distance is not taken
as a "square", instead the algebraic sign of the distance seems to be
of some importance. My results were quite different, whenever I took
(y[i]-Yi) instead of (Yi-y[i]) (as it is shown in the example).

Taking the latter the solver works fine, taking the first distance the
results have been quite funny.

Andreas Gros 


Andreas Gros, Tel.: 0179-1062416
Lotter Str. 99, 49078 Osnabrück
Amalienstr. 8, 97072 Würzburg
http://www.gros-webdesign.de