[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Help-glpk] Re: How to avoid output when reading in an mps problem.

From: Enea Zaffanella
Subject: Re: [Help-glpk] Re: How to avoid output when reading in an mps problem.
Date: Thu, 27 Sep 2007 17:02:56 +0400

Andrew Makhorin wrote:
>> In our software project (the Parma Polyhedra Library,
>> we use glpk in a demo application, mainly
>> for debugging purposes: we have developed an exact arithmetic mip solver
>> and we want to check if the results we are obtaining agree with those
>> obtained using glpk.
> That is very interesting. Probably it may be interesting to you that
> glpk has an exact lp solver (lpx_exact) based on rational arithmetic,
> which is also available in glpsol (options --xcheck and --exact).
> However, this is an exotic feature (I mean glpk), because instances
> of real size cannot be solved/checked in a resonable time.

I plainly admit that I wasn't aware of the availability of an exact lp 
solver in glpk: I will certainly have a look at it.

As for usability of such a solver ... it mainly depends on the context.
We are not addressing "real size" problem, if by that you mean the 
typical size of an LP/MIP problem specified for optimization purposes. 
In our context, which is static analysis and verification of 
hardware/software systems, we need to obtain exact results for what you 
may consider to be (many) "small" problems. Actually, we could even 
accept approximated results, provided that all the approximations are 
done on the safe side only ... which in the end means that exact 
arithmetic is needed anyway. Also, we sometimes need to solve many small 
variations of the same problem, so that a support for the incremental 
addition (or removal) of constraints is something we value positively.

Thanks again,
Enea Zaffanella.

>>  It is acceptable that the user hasn't installed
>> glpk at all, or that an old version is found. Hence, it is not really
>> appropriate to have our project depend on glpk. Rather, at configure
>> time, we try to detect the installed version of glpk and, based on that,
>> we conditionally enable/disable the various tests.
>> With the valuable info you provided we should be able to write 
>> meaningful auto-configuration code.
> Starting from glpk 4.16 there are two preprocessor variables
> GLP_MAJOR_VERSION and GLP_MINOR_VERSION defined in glpk.h as well as
> the api routine glp_version which may be used to check the version
> of the package.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]