Re: [Help-glpk] Black box Solver

[Brady Hunsaker]

Andrew (Whyte),

The way that the underlying algorithms work, it's not generally enough 
to just have a "black-box" objective value answer.  The algorithms 
expect that formulations have linear objectives, which allows them to 
have derivative information to direct them to better solutions.

If the objective function is not linear, then you need a different 
approach; one that isn't supported by GLPK.  You may want to look into 
using some kind of local search approach.

Brady

Andrew Whyte wrote:
> Hi!
>
> I'm trying to implement the GLPK simplex solver into an application,
> however my objective function isn't expressible in a suitable form for
> the input.  
> Instead, as the gnumeric team have done, I'd like to treat the objective
> function as a black box that supplies a number to GLPK based on the
> coefficients that GLPK requests.  I assume an iterative procedure is
> used until GLPK returns a "solution found" code.
> I've tried to work through the gnumeric code but it has been abstracted
> to work with lp_solver as well - so I'm finding it very hard to figure
> out how it actually works!
> Could anyone explain which API calls are involved (I'm quite new to
> linear programming but I've worked through quite a lot of documentation
> on the basic principles, and the GLPK RefMan).
>
> Thanks in advance,
>
> Andrew
>
> ps. I'll be using the JNI implmentation to talk to GLPK but I think all
> of the API calls are included so that (hopefully) shouldn't matter...
>
>
>
> _______________________________________________
> Help-glpk mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/help-glpk


--
Brady Hunsaker
Assistant Professor
Industrial Engineering
University of Pittsburgh
http://www.engr.pitt.edu/hunsaker/