[Help-glpk] reduced cost fixing

[andre camatta]

Hi,
        I am wondering whether any variable fixing operation
has been done 
at the nodes of branch and bound mip solver. Or it's
just a normal 
branch 
and bound framework based on the lp solver. How the
branching variables 
are chosen in this implementation?

thanks,
-shubhashis

Reduced cost fixing (IOS framework, only Binary MIP):
#include "glpk.h"

void reduced_cost_fixing (IOS *ios)
{
        int i, n, tagx;
        double zi, vx, dx, tol = 1e-6;
        n = ios_get_num_cols (ios);
        zi = ios->lp_obj;
        for (i = 1; i <= n; i++)
        {
                tagx = ios_get_col_soln(ios, i, &vx , &dx);
                if (tagx != IET_BS && tagx != IET_NS &&
ios_get_col_kind(ios, i) == IOS_INT)
                {
                        if (vx < tol && vx > -tol && ios->best - zi < dx)
                        {
                                ios_set_col_bnds(ios, i, IOS_FX, 0.0, 0.0);
                        }
                        else if (vx < 1 + tol && vx > 1 - tol && zi -
ios->best > dx)
                        {
                                ios_set_col_bnds(ios, i, IOS_FX, 1.0, 1.0);
                        }
                }
        }
        ios->p_stat == IOS_FEAS;
        ios->d_stat == IOS_FEAS;
        return;
}
 
In the application procedure:
switch (ios->event)
{  
...
    case IOS_V_BRANCH:
    if (ios->found == 1)
        reduced_cost_fixing (ios);
    branching (ios);
...
}

My procedure need improvements and tests.

André Camatta           

______________________________________________________________________

Yahoo! GeoCities: 15MB de espaço grátis para criar seu web site!
http://br.geocities.yahoo.com/