[Scip] Presolving Issues

michael.winkler@zib.de michael.winkler at zib.de
Sun Mar 10 14:40:43 MET 2013


Hi Andreas,

theer are different things you can try to fasten presolving. Without
seeing the statistics (in the interactive shell you can execute "display
stat" after you presolved your instance) it is not clear where the time
went.
But here are some setting ideas, what you can do:

1. set pres emph fast
2. set pres domcol maxrounds 0
3. set pres compo maxrounds 0

The first should be the most powerful in form of fasten your presolving.
You can also try to combine the second and third. (Using the first/maybe
also the one of the other setting you probably do not need to use your
round anymore.)

(If you are not using the first setting you can disable probing by "set
prop prob maxpre 0".)

Can you also provide us with your instance.

Best, Micha

> Hi all,
>
> I have some issues with the presolving settings.
> The IP I'm working with has a lot of useless variables and constraints.
> Presolving deletes them in the first rounds and goes on for a while
> without breath-taking results.
> So I tried to set maxrounds to 10.
> It seems as if this is finished within 5 seconds but it lasts another 55
> seconds until the solving process is started (I pasted the output at the
> bottom of the mail).
> Are there any settings I could use to avoid this behaviour?
>
> I hope, some of you guys are working on a sunday afternoon as my project
> deadline is tonight.
>
> Thanks a lot!
> Best regards,
> Andreas
>
> output:
> presolving:
> (round 1) 188689 del vars, 288346 del conss, 0 add conss, 208160 chg
> bounds, 11470 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 2) 200195 del vars, 299852 del conss, 0 add conss, 208160 chg
> bounds, 13453 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 3) 200756 del vars, 299853 del conss, 0 add conss, 208160 chg
> bounds, 13453 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 4) 200756 del vars, 300113 del conss, 0 add conss, 208514 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 5) 201019 del vars, 300437 del conss, 0 add conss, 208514 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 6) 201080 del vars, 300437 del conss, 0 add conss, 214135 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 7) 201103 del vars, 300437 del conss, 0 add conss, 214138 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 8) 201103 del vars, 300452 del conss, 0 add conss, 214138 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
> (round 9) 201103 del vars, 300479 del conss, 0 add conss, 214154 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4162 impls, 32 clqs
>    (2.9s) probing: 80/330 (24.2%) - 0 fixings, 0 aggregations, 54
> implications, 0 bound changes
>    (2.9s) probing aborted: 50/50 successive totally useless probings
> (round 10) 201104 del vars, 300480 del conss, 0 add conss, 214178 chg
> bounds, 13655 chg sides, 0 chg coeffs, 0 upgd conss, 4252 impls, 32 clqs
>    (4.7s) probing cycle finished: starting next cycle
>    (4.8s) probing: 51/330 (15.5%) - 0 fixings, 0 aggregations, 414
> implications, 0 bound changes
>    (4.8s) probing aborted: 50/50 successive totally useless probings
> presolving (11 rounds):
>  201261 deleted vars, 300619 deleted constraints, 0 added constraints,
> 215210 tightened bounds, 0 added holes, 13832 changed sides, 0 changed
> coefficients
>  4743 implications, 32 cliques
> presolved problem has 32162 variables (330 bin, 31832 int, 0 impl, 0 cont)
> and 15557 constraints
>      77 constraints of type <varbound>
>      32 constraints of type <setppc>
>   15412 constraints of type <linear>
>      36 constraints of type <logicor>
> transformed objective value is always integral (scale: 1)
> Presolving Time: 60.10
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
>



More information about the Scip mailing list