[Coin-lpsolver] CLP and Matlab, perfect together?

[Mikhail Nediak]

Just a thought...

I guess it would be much better to implement MATLAB interface for a
general LP solver supporting OsiSolverInterface. In this case you can
link it with any solver that already has OsiSolverInterface (CPLEX, OSL,
CLP, GLPK, ...). Last time I looked, I do not remember seeing a method in
OsiSolverInterface that would let you select an entering rule. But then
this would be the only extra "hack" you would have to add (this is instead
of implementing a customized interface to an individual solver, such as
CLP, whose own interface might change, which would require changes in
your code, etc).

Best,
Mikhail Nediak

On Thu, 16 Jan 2003, David de la Nuez wrote:

> Hello friends and colleagues!  I am PhD student of Don Goldfarb at Columbia
> University in the City of New York, and a brand-new member of this
> discussion list.  We are considering replacing cplex as an LP solver in our
> work, for reasons I will spare you at the moment.
>
> The issue at hand is that I live in Matlab, so to speak.  While I have
> coded in several different languages / environments, so many things are so
> easy in Matlab that the early decision was made to use it (other
> complications I will also spare you).  I had been using Mikael Prytz's
> "cplex mex"  Matlab interface for cplex 6.5 for a number of months and
> modified it (open source is the best!) to work with cplex 8.
>
> I would be thrilled to use an open source LP solver which is known to be of
> industrial strength.  I am aware of GNU's LP package, but shied away from
> it, not having any reference for how good it was.  Enter COIN LP -- it
> seems the ideal solution.  However, the dearth of documentation and my lack
> of familiarity with C++ (have not used it for more than four years!) cast
> doubt on my belief that I could put together a Matlab interface for CLP in
> a reasonable amount of time.
>
> I suppose, then, after the long introduction above, that I would like to
> know if anyone in the community could help me out in creating a Matlab
> interface for CLP, or at least a raw version I could gradually improve on
> my own.  At some point in the future a move to pure C/C++ would make a lot
> of sense for my project, but now is not the time (yet another long story).
>
> The functionality I need is not too much.  I need to be able to create an
> LP, solve it, add columns, solve it, etc -- I am using some column
> generation, as you might have guessed.  I need to be able to manipulate the
> entering rule (i.e. use Dantzig and steepest edge or some other rule) and
> some options like number of iterations, etc.  It would be great if someone
> could help me get started or at least if we could get a discussion going.
>
> By the way, it appears that Prytz's "cplex mex" is no longer available
> anywhere (at least not via a Google search).  If anyone would like it
> and/or my "updated" hacked version for cplex 8, let me know.
>
> P.S. Anyone from New Jersey might "get" the subject header of my
> email.  For those who don't, "New Jersey and you, perfect together" was the
> state slogan years ago.
>
> David de la Nuez
> dmd57 at columbia.edu
> http://www.ieor.columbia.edu/~oboeguy
>
> "The object of education is not the production of self-confident fools."
> 	- Jerome Bruner
>
> _______________________________________________
> Coin-lpsolver mailing list
> Coin-lpsolver at www-124.ibm.com
> http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-lpsolver
>