Relaxation vs Root solve

737 views
Skip to first unread message

Luis Pinto

unread,
May 30, 2014, 10:19:09 AM5/30/14
to AIMMS GoogleGroups
Hello All,

Need some help with tuning a model.

My problem is a MIP with ~400k Variables (8k integers) and 45k Constraints.

What I'm trying to figure out is why there is such a big different between solving the root node and solving the relaxed model.

I explain:

When I solve the model relaxed (setting the "Relax status" for these variables) the model runs in ~3 minutes.

When I solve the normal model (and check the .sta file generated by CPLEX) I find that after 20 minutes the model hasn't even finished the Root node processing.
I know this because no time has gone into the "b&c" phase.

I have explicitly set the LP method to Primal for both LP and Starting method for MIP, so that can't be the problem.

I also tried to eliminate the presolve (in hope that the MIP presolve is just taking longer), but no success either (relaxed took longer, but MIP still can't solve the root).

Any ideas as too the difference between them and how I can improve the root solve?

Regards,

Luis Pinto

www.unisoma.com

Marcel Hunting

unread,
May 30, 2014, 11:01:26 AM5/30/14
to ai...@googlegroups.com
Hi Luis,

Normally the CPLEX status file will show a line

   Root relaxation solution time = ...

and the time shown there should be about equal to the solving time of your relaxed model. Is that the case?

After solving the root relaxation, CPLEX might do extra things before branching, including:
  1. Adding cuts, resolve root node (and this can be repeated several times);
  2. Running heuristics to find good incumbent solution.
These extra things can consume a lot of time (but they in general pay off during the branch-and-bound phase).

Best regards,

Marcel Hunting
AIMMS

Luis Pinto

unread,
May 30, 2014, 11:07:16 AM5/30/14
to AIMMS GoogleGroups
Thats excatly it, for the relaxed version, I have the solution time in ~3minute.
For the MIP, the "Root relaxation solution time" is the time limit establiash (in other words 20min).


--
You received this message because you are subscribed to the Google Groups "AIMMS - The Modeling System" group.
To unsubscribe from this group and stop receiving emails from it, send an email to aimms+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Marcel Hunting

unread,
May 30, 2014, 11:52:12 AM5/30/14
to
Maybe you can set the CPLEX option 'Simplex Display' to 'At refactorizations' or 'Every iteration' (but the status file might become large with the latter setting) and the CPLEX option 'MIP Display' to 'Nth node + LP display for root node', run the MIP and the RMIP again, and then check the differences in the status file? If that does not give a clue, can you then post the resulting status files here or send it to AIMMS support?

Marcel

Luis Pinto

unread,
Jun 6, 2014, 8:13:19 AM6/6/14
to AIMMS GoogleGroups
Hey Marcel,

I'm still trying to figure it out. But for now we were able to workaround.

Cheers,


On 30 May 2014 12:52, Marcel Hunting <Marcel....@aimms.com> wrote:
Maybe you can set the CPLEX option 'Simplex Display' to 'At refactorizations' or 'Every iteration' (but the status file might become large with the latter setting) and the CPLEX option 'MIP Display' to 'Nth node + LP display for root node', run the MIP and the RMIP again, and then check the differences in the status file? If that does not give a clue, can you then post the resulting status files here or send it to AIMMS support?

Marcel
Reply all
Reply to author
Forward
0 new messages