Function interface for Least-Squares Fitting

Function interface for Least-Squares Fitting

[Jari Häkkinen]

Hi,

I noticed that the function interface for the fitting algorithms do not 
support gsl_vectors. What is the reason for this? Would it be possible 
to add inline functions such as

inline int
gsl_fit_linear_vector (const vector * x, const vector * y
                        double * c0, double * c1,
                        double * cov00, double * cov01, double * cov11,
                        double * sumsq)
{
   return gsl_fit_linear(x->data, x->stride, y->data, y->stride, x->size,
                         c0,c1,cov00,c0v01,cov11,sumsq);
}

to GSL?


Jari

Re: Function interface for Least-Squares Fitting

[Brian Gough]

Jari Häkkinen writes:
 > I noticed that the function interface for the fitting algorithms do not 
 > support gsl_vectors. What is the reason for this?

Basically to avoid duplication, when possible.  

The recommended way to call these functions is with v->data,
v->stride, v->size as in your example. 

You can put an inline function definition in your application's own
header files if you like.

-- 
Brian Gough

Re: Function interface for Least-Squares Fitting

[Jari Häkkinen]

Brian Gough wrote:

> Jari Häkkinen writes:
>  > I noticed that the function interface for the fitting algorithms do not 
>  > support gsl_vectors. What is the reason for this?
> 
> Basically to avoid duplication, when possible.  
> 
> The recommended way to call these functions is with v->data,
> v->stride, v->size as in your example. 
> 
> You can put an inline function definition in your application's own
> header files if you like.

Yes, of course, I am doing it my copy of the gsl CVS (in cvs_fit.h) but 
would like to see it the official release. Why was it decided against 
having gsl_vectors in the interface. (I am maintaing a C++ interface to 
GSL, and in this interface there is no access to internals of gsl_vector.)


Jari

Re: Function interface for Least-Squares Fitting

[Brian Gough]

Jari Häkkinen writes:
 > Yes, of course, I am doing it my copy of the gsl CVS (in cvs_fit.h) but 
 > would like to see it the official release. Why was it decided against 
 > having gsl_vectors in the interface.

This is to avoid having additional dependencies or two versions of
every function---where possible, the plain C interface is preferred.
See the GSL Design document for more explanation.

-- 
best regards

Brian Gough

Network Theory Ltd,
Commercial support for GSL --- http://www.network-theory.co.uk/gsl/