Re: [Help-gsl] Linear least squares, webpages and the next release

[Foivos Diakogiannis]

Hi to all,

1. Is anyone out there using GSL besides me? :-)

A *lot* of people in the astronomy/astrophysics community, it is rather
hard to list everyone. It should also not be overlooked how important is
the source code of GSL: I've learnt a lot of things for good programming
techniques from GSL code.

2. What functionality would you like to see added to GSL?

a. Boundary value problems for partial differential equations.
b. I know it is too much to ask, but I'd love to see a c++ implementation
of GSL (not just a wrapper), something like Eigen/boost - style.
c. Markov Chain Monte Carlo parameter estimation, with various methods
(e.g. Hamiltonian monte carlo, Population mcmc etc).
d. I am not happy with the performance of the numerical integration
routines, they are generally way too slow (by the way, I've never managed
to make the Gauss-Legendre quadrature to work properly). The most reliable
I've found is the CQUAD. In comparison, see
http://www.holoborodko.com/pavel/numerical-methods/numerical-integration/
 from Pavel Holoborodko for fast GL quadratures. Again for comparison only,
Numerical recipes is also a lot faster (not as reliable though).

3. Are you willing to develop and contribute the features you want?

I am afraid I do not meet the standards of GSL, and I do not have time to
study and become proficient in C - style. Am more than happy to contribute
in C++ style code. I have a b-spline c++ class, and based on that some
b-spline solvers for a restricted class of boundary value problems. I am
also preparing (slowly) an mcmc implementation for population mcmc methods.


4. Would you like to see a quick release of GSL v2.0, or are you content to
work off the git repository?

yes

Regards,
Foivos


On Fri, Oct 23, 2015 at 3:17 AM, Patrick Alken <address@hidden>
wrote:

> Hi all,
>
>   There is significant new capability in GSL now for regularized linear
> least squares problems. I've added routines to do L-curve analysis as well
> as handle non-square Tikhonov regularization matrices. For those of you who
> have large least squares problems (ie: millions of rows in your matrix)
> stay tuned, as I'm working on adding some functionality for such problems.
>
>   I've updated the GSL webpages on gnu.org to the latest in the git
> repository, since many new features have been added since the last 1.16
> release. The main page also now recommends to install GSL from the git,
> since 1.16 is quite outdated. This brings me again to the discussion of the
> next release of GSL.
>
>   The last discussion of GSL v2.0 was in June 2014 (time goes too
> quickly!). I would like to reopen this discussion. The main points back
> then holding up v2.0 were:
>
> 1. New memory management (redesign of all GSL containers)
> 2. Incorporating LAPACK/LAPACKE/libflame
>
> As far as I know, nothing has been done on either of these points, however
> a great deal of new functionality has been added to GSL in the meantime
> (sparse matrices, regularized least squares, running statistics, 2D
> interpolation, monotonic interpolation, new nonlinear least squares
> solver). I think (as I stated last year) that points 1 and 2 should be
> postponed for GSL v3.0, and we should release v2.0 soon. At this point I am
> willing to make an executive decision and just do it :-) but I would like
> to check in with everyone first.
>
> Over the past 2 years or so, looking at the git logs, I've been (mainly)
> the only person adding code to GSL. So I'd like to ask some poll questions
> (and recruit talented people to work on things) and so I'm posting to
> help-gsl as well. Here are some questions:
>
> 1. Is anyone out there using GSL besides me? :-)
>
> 2. What functionality would you like to see added to GSL?
>
> 3. Are you willing to develop and contribute the features you want?
>
> 4. Would you like to see a quick release of GSL v2.0, or are you content
> to work off the git repository?
>
> Thanks,
> Patrick
>
>