Re: [Help-gsl] GSL 1.9: Levenberg-Marquardt non-linear fit with selected invariant coefficients not to be fit, HOWTO?

[Oliver Jennrich]

On 8/1/07, O. Hartmann <address@hidden> wrote:
>
> Hello,
>
> while using GSL the first time (coming from some Numerical Recipes in C
> code pieces) I ran into trouble solving a non-linear least square problem.
> The problem ist simple, not to say trivial, some points need to get
> fitted in the usual way. The modell-function is a polynomial expression
> of degree 15 with 16 coefficients (a_0 - a_15) and in most cases we only
> fit the first coefficient, the constant (a_0). Sometimes we need to
> select some other coefficients to be fitted for making the polynomial
> fitting the dataset.
> Well, The Numerical Recipes code offers an elegant method using a 'mask'
> for each coefficient, if the corresponding switch in the mask (the mask
> represents the amount of coefficients) is set to 'false' or 0, the
> coefficient does not get touched and fitted, if set to else, the
> coefficient is handled as usual and gets fitted. These operations are
> done while preparing the covariant matrix.
> The GSL routines, as shown in the example in chapter 37: Nonlinear
> Least-Square Fitting seems not to offer a simple 'switch' to do the same
> as mentioned above. Does anyone do have a hint?

Well, you can pass whatever data you want to the function through the
'data'  parameter. It is declared 'void'  so you are responsible to
extract the right data. Define a struct

 struct data {
       size_t n;
       double * y;
       int * mask;  /* 1 or 0 */
     };

and use the 'mask'  to multiply the coefficients...

-- 
Space -- the final frontier