Re: [Help-glpk] [Fwd: Can MathProg Translator generate C or C++ code snippet?]

[Harley Mackenzie]

This certainly would be technically feasible and a very useful addition 
to the GLPK program, however, it would take a lot of time to develop and 
would require someone who has a lot of motivation to implement.

The benefit for all users would be that you could develop the model in 
MathProg and, once debugged and refined, generate a C++ program for 
repeated runs and to allow greater output flexibility and control of the 
running of the program.

Another alternative would be to output the model in some simple to load 
representation of the matrix elements (JSON, XML, binary?) so that it 
could then be loaded quickly into a C/C++ GLPK program.

Anyone interested in this sub-project?

Regards,

Harley

On 02/09/2012 12:23 AM, Haroldo Gambini Santos wrote:
> On 08-02-2012 12:12, Andrew Makhorin wrote:
> > -------- Forwarded Message --------
> > From: Sebastian Kranz<address@hidden>
> > To: address@hidden
> > Subject: Can MathProg Translator generate C or C++ code snippet?
> > Date: Wed, 08 Feb 2012 10:57:02 +0100
> >
> > Dear Developers,
> >
> > I find the possibility of GLPK to read and process
> > MathProg models extremely helpful. Thanks a lot for
> > creating the translator!
> >
> > In my view, MathProg is usually a much more convenient
> > form to specify a linear program than directly filling the
> > constraint matrix with numbers. However, I have an
> > application that has to solve the same model many times
> > with different specifications of the sets and the
> > parameters of the MathProg model. In this application,
> > creating every time a new instance via the MathProg
> > translator seems rather slow in terms of computing speed.
> >
> > I wondered whether it would be possible to add a function
> > to the MathProg translator that can create from the
> > internal model representation of a given MathProg model
> > some C code snippet. The snippet would be a C (or C++)
> > function that takes sets and parameters as vector input
> > and returns the constraint matrix, bounds and objective
> > coefficients of the linear program as output. I could
> > imagine that this could be quite useful, since one could
> > specify the logic of the problem in the convenient
> > MathProg form, but still can embed in the application fast
> > code to generate the linear programs.
> Usually the slow part of the process is not generating the model but
> solving it - specially for integer programs.
>
> As a GLPK user I'm much more interested in improvements in the
> robustness of the liner programming solver and in the algorithms for
> integer programming than some speedup in generating the model.
>
> GLPK is still one of the slowest solvers:
> http://plato.asu.edu/ftp/milpc.html
>
> There is a long way to go for improving the solver code.
>
> Cheers,
>
> Haroldo
>
> > However, I have no clue whether that is possible or hard
> > it would be to create such a snippet generator.
> >
> > Best wishes,
> > Sebastian
> >
> >
> >
> > _______________________________________________
> > Help-glpk mailing list
> > address@hidden
> > https://lists.gnu.org/mailman/listinfo/help-glpk

--
------------------------------------------------------------------
     Dr. Harley Mackenzie         ABN:   36 348 783 012

     HARD Software                Web:   www.hardsoftware.com
     PO BOX 8004                  Tel:   +61 3 5222 3435
     Newtown 3220, Australia      Email: address@hidden
------------------------------------------------------------------