An even more ambitions project than GSL, would be to try and tap into PETSc 
(Portable, Extensible Toolkit for Scientific Computation)!

PETSc provides a nice collection of solvers, matrix preconditioners etc., 
beyond what Lapack2 offers and presumably much higher performance than calling 
R through RServer.


Unfortunately this appears more complex than simply calling a DLL: typical 
usage is apparently to write code in C and use the PETSc compiler mpicc which 
hides most of the uggly linking issues from the C backend compiler.  PETSc can 
parallelize calculations over multiple processors, using OpenMPI for 
communications; I guess this might be hard to exploit in J?  Python has 
bindings to PETSc, so somehow something should be possible...



Maybe a reasonable starting point would be to export matrices to HDF5 or netCDF 
(via R or using text and command line tools to start off with), and prepare the 
PETSc transformations in C or Python :-(, in  a strongly decoupled way.

But it would be so much nicer to be able to be able to do it all from J (even 
if this means using a single thread).  Maybe it could even convert some users 
to J.


De : Scott Locklin <[email protected]>
À : [email protected]
Sujet : Re: [Jprogramming] GNU Scientific bindings?
Date : 01/03/2021 21:19:53 Europe/Paris

Kind of curious: what is compelling about GSL? I ported a few things from
it to a Lisp that had partially completed bindings, mostly because it was
already basically there, but it's generally a trashfire of very little
utility to a language like J which already has most of it built in.

For example, while I am not aware of any simulated annealing J gizmoes,
differential evolution is almost always a better algorithm for
metaheuristic optimizers and it is implemented here: math/deoptim . It
worked the last time I checked! Was pretty fast too.

-SL


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to