Re: [Help-glpk] Setting Up Gap Tolerance in Gusek

[Seung Mo Gu]

Oh, so I guess then there really is no "gap" per say. I was comparing my 
current best bound with the lp relaxation objective value. Thank you for 
clearing this up. It turns out the code at the time still had no feasible 
solution even after running for 60 hours.


After modifying my model a bit, a few feasible solutions were found. I guess 
the question remains how one would go about accepting the current best solution 
when a certain gap is achieved. There seems to be many references made to 
glp_intopt but as I am new to GLPK/GUSEK I am confused on how to do this.

On Wed, Apr 7, 2010 at 12:24 PM, Andrew Makhorin <address@hidden> wrote:
> I am currently trying to set up a gap tolerance in GUSEK for my
> MIP. The solver manages to get within a decent gap within a reasonable
> amount of time and when stopped, GUSEK terminates without displaying
> any solution. Is there a way to have the GUSEK stop and accept a
> feasible solution as a valid result in the case that it reaches a gap


> of let #39;s say 5%?

Please provide the glpsol output that appears in the output panel.

Note that you can run glpsol directly from the output panel, i.e. from
the command line, specifying necessary options to save the solution.

> Also, one other thing I noticed when running the solver is that the


> obj value doesn #39;t always decrease for a minimization problem. Does
> the status window display all feasible (infeas isn #39;t zero but
> close to it i.e. 10^-14) solutions or should it only update when a new
> best solution is found?



If solution of lp relaxation takes more than 5 secs, you can see
output from the dual simplex, which is indicated by vertical bar and
looks like follows:

|     4: obj =   7.333333333e+02  infeas =  0.000e+00 (0)

The objective value displayed is related to the lp relaxation being
solved, and in case of minimization that value increases, because the
dual simplex is used.








-- 
Seung Mo Gu

Undergraduate Student
Georgia Institute of Technology
ISyE Operations Research and Statistics




Oh, so I guess then there really is no "gap" per say. I was comparing my current best bound with the lp relaxation objective value. Thank you for clearing this up. It turns out the code at the time still had no feasible solution even after running for 60 hours.

After modifying my model a bit, a few feasible solutions were found. I guess the question remains how one would go about accepting the current best solution when a certain gap is achieved. There seems to be many references made to glp_intopt but as I am new to GLPK/GUSEK I am confused on how to do this.

On Wed, Apr 7, 2010 at 12:24 PM, Andrew Makhorin <address@hidden> wrote:

> I am currently trying to set up a gap tolerance in GUSEK for my
> MIP. The solver manages to get within a decent gap within a reasonable
> amount of time and when stopped, GUSEK terminates without displaying
> any solution. Is there a way to have the GUSEK stop and accept a
> feasible solution as a valid result in the case that it reaches a gap

> of let #39;s say 5%?

Please provide the glpsol output that appears in the output panel.

Note that you can run glpsol directly from the output panel, i.e. from
the command line, specifying necessary options to save the solution.

> Also, one other thing I noticed when running the solver is that the

> obj value doesn #39;t always decrease for a minimization problem. Does
> the status window display all feasible (infeas isn #39;t zero but

> close to it i.e. 10^-14) solutions or should it only update when a new
> best solution is found?

If solution of lp relaxation takes more than 5 secs, you can see
output from the dual simplex, which is indicated by vertical bar and
looks like follows:

|     4: obj =   7.333333333e+02  infeas =  0.000e+00 (0)

The objective value displayed is related to the lp relaxation being
solved, and in case of minimization that value increases, because the
dual simplex is used.

--
Seung Mo Gu

Undergraduate Student
Georgia Institute of Technology
ISyE Operations Research and Statistics