Statistical tests

Statistical tests

[Alberto RIBON]

  Hello,

  I am not writing for reporting a bug, but for
  asking a question: I have looked through the various
  GSL functions but I haven't found any function
  related to statistical tests, that is to determine
  the probability that a distribution (histogram) is
  coming from a certain distribution (which could be  
  an analytical function, or another histogram).
  Examples of such tests are:
  
   --- Kolmogorov-Smirnov test;
   --- Smirnov-Cramer-von Mises test;
   --- Anderson-Darling test.

  Am I right, or these functions are already provided
  in GSL? If not, as I think, is it planned to have 
  them in future extensions, and if this is the case
  when more or less?

   Thanks,
            Best regards,
                                 
                           Alberto Ribon.

Re: Statistical tests

[Brian Gough]

Alberto RIBON writes:
 >   I am not writing for reporting a bug, but for asking a question:
 >   I have looked through the various GSL functions but I haven't
 >   found any function related to statistical tests, that is to
 >   determine the probability that a distribution (histogram) is
 >   coming from a certain distribution (which could be an analytical
 >   function, or another histogram).  Examples of such tests are:
 >    --- Kolmogorov-Smirnov test; --- Smirnov-Cramer-von Mises test;
 >    --- Anderson-Darling test.
 >   Am I right, or these functions are already provided in GSL? If
 >   not, as I think, is it planned to have them in future extensions,
 >   and if this is the case when more or less?

GSL doesn't have any statistical tests at the moment, since nobody has
volunteered to write them.  If you'd like to write some let me know.

Brian

Re: Statistical tests

[Alberto Ribon]

  Hello,

  I am involved, with other people (mainly from CERN,
  European Organization for Nuclear Research, 
  and also from INFN, the Italian National Nuclear Physics
  institute), in a project to develop a statistical test suite 
  to automatize the tedious but essential task of checking
  a large (thousands) distributions (either from real experimental
  data or from simulation), against other "reference" distributions
  (which can be theoretical expected distributions, or other
   data distributions, or even simulated distributions, obtained
   for example in previous versions of the simulated package).

  Central in this project is the employment of different 
  statistical tests, like Chi2, Kolmogorov-Smirnov, 
  Smirnov-Cramer-von Mises, Anderson-Darling, etc...
 
  This project will be written in  * C++ *  and will be
  * free software *. Furthermore, for the histogram part,
  we plan to use another C++ free software package, developed
  in the same High Energy Physics community, called AIDA/Anaphe:

      http://aida.freehep.org/

      http://cern.ch/anaphe/

  (another package, still in C++ and free software, which is
   currently heavily used by all High-Energy Physics community,
   and which provides lots of useful mathematical functions, is
   CLHEP:

      http://cern.ch/clhep/
 
   this is not directly linked with the statistical tests, but
   I think it is worth to bring it to your attention)

   We plan to have a first release of it for Spring-Summer 2003.

   We are quite interested and pleased to provide this code
   to GSL, including the necessary integration effort with
   the rest of GSL. 

   We will contact you as soon as we have a working first version.
  

       Best Regards,

                               Alberto Ribon


   PS- I'd like to thank and acknowledge for the 
       replies I received previously (see below), from Ramon Diaz
       and Jason H. Stover, who pointed me on the GNU R project.
       I had a look at it, but we don't think it provides what
       we really need.


-------------------------------
On Mon, 9 Dec 2002, Brian Gough wrote:

->Alberto RIBON writes:
-> >   I am not writing for reporting a bug, but for asking a question:
-> >   I have looked through the various GSL functions but I haven't
-> >   found any function related to statistical tests, that is to
-> >   determine the probability that a distribution (histogram) is
-> >   coming from a certain distribution (which could be an analytical
-> >   function, or another histogram).  Examples of such tests are:
-> >    --- Kolmogorov-Smirnov test; --- Smirnov-Cramer-von Mises test;
-> >    --- Anderson-Darling test.
-> >   Am I right, or these functions are already provided in GSL? If
-> >   not, as I think, is it planned to have them in future extensions,
-> >   and if this is the case when more or less?
->
->GSL doesn't have any statistical tests at the moment, since nobody has
->volunteered to write them.  If you'd like to write some let me know.
->
->Brian
->

Addendum to previous email

[Alberto Ribon]

  Hello,

  after talking with my collaborators, we confirm our
  interest in providing our contributions to GSL as far as
  the statistical tests are concerned, but for the integration
  effort we would appreciate the help of experts in GSL and
  any commitments from our side would be obviously subject to
  an evaluation of the effort needed.

  By the way, we would appreciate any information of guidelines
  and possible constraints relevant for the integration of code in GSL.

     Best Regards,
                               Alberto

Re: Addendum to previous email

[Jason H. Stover]

I'm willing to help develop and integrate statistical
tests, though right now I'm buried under cumulative 
distribution functions.

-Jason

On Tue, Dec 10, 2002 at 01:11:15PM +0100, Alberto Ribon wrote:
> 
>   Hello,
> 
>   after talking with my collaborators, we confirm our
>   interest in providing our contributions to GSL as far as
>   the statistical tests are concerned, but for the integration
>   effort we would appreciate the help of experts in GSL and
>   any commitments from our side would be obviously subject to
>   an evaluation of the effort needed.
> 
>   By the way, we would appreciate any information of guidelines
>   and possible constraints relevant for the integration of code in GSL.
> 
>      Best Regards,
>                                Alberto
> 
> 
> 

Re: Addendum to previous email

[Brian Gough]

Alberto Ribon writes:
 >   By the way, we would appreciate any information of guidelines
 >   and possible constraints relevant for the integration of code in GSL.

The 'GSL Design Document', available from http://sources.redhat.com/gsl/