Re: [Help-glpk] Column Generation

[Marc Goetschalckx]

On 12-Mar-14 12:38, Andrew Makhorin wrote:
> > I am trying to develop a column generation algorithm for a MIP problem
> Please note that glpk mip solver does *not* support column generation.
> Only rows (lazy constraints and/or cutting planes) can be added to
> subproblems during the search.
>
> >   and I want to calculate the reduced costs of my variables.
> >
> > I do not have access to the API but I can only call the glpsol
> > executable and so I thought I would  use the .dual suffix after
> > “--read"ing the solution of the current subproblem.
> >
> > The problem is that the .dual function always returns 0, for every
> > structural variable.
> >
> > I thought that the problem was I have a MIP problem instead of a LP
> > problem. So I ran glpsol with the --nomip option but the result is
> > still 0!
> >
> > What am I doing wrong?
> If your mathprog model has integer variables, it is considered as mip
> independently on --nomip option. The latter affects only output from the
> problem object (e.g. -o option) and doesn't affect which solution
> components (lp or mip) go into the model. And since for mip solution
> dual variables have no meaning, they all are set to zero.

The --nomip option states that it considers all variables as continuous 
variables.

Options specific to MIP solver:
   --nomip           consider all integer variables as continuous
                     (allows solving MIP as pure LP)

I assumed it solved the linear relaxation of the (mixed-)integer problem.
That is significantly different from the explanation provided above by 
Andrew.
Dual variable values should be available at the conclusion of the 
solution algorithm.
If that is not the case, this is a bug report since the program behavior 
is inconsistent with the documentation.
> _______________________________________________
> Help-glpk mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/help-glpk

--
Marc Goetschalckx