[SCIP] experiment with primal heuristics separately

[Gregor Hendel]

Dear Ahmad and list,

it is possible via SCIP settings to disable almost everything 
(presolving, separation) except for a certain heuristic in the solution 
process, and set the node limit of this solution process to 1. You may 
need to adjust the frequency and frequency offset parameters of the 
heuristic to achieve it running at the root node. E.g., to only run 
simple rounding, which requires an LP to be solved, you may call

set presol emph off
set sepa emph off
set branch random prio 1000000
set lim nodes 1
set heuristics emph off
set heur simplerounding freq 1
read check/instances/MIP/bell5.mps
opt
disp stat
quit

The statistics will show you that not much more than an LP solve and 
simple rounding were executed. By replacing simplerounding with the name 
of the heuristic you want to run, you basically get the solving process 
to do what you intend.

Note that in general, the execution of primal heuristics is connected to 
particular events during the search. Also, the working limits for the 
primal heuristics are determined regarding statistics of the overall 
solution process (elapsed search nodes, elapsed LP iterations).

Cheers,
Gregor


Am 23.12.2016 um 01:26 schrieb Ahmad Moradi:
> Dear SCIPers,
>
> I would only apply some of the primal heuristics implemented by SCIP 
> to study the quality of found feasible solutions and I don't actually 
> need to solve them to optimality through branch and bound. So,
>
> Is it possible to _/run/_ a primal heuristic of SCIP on a MIP instance 
> without solving the MIP?
>
> Best regards,
> Ahmad
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20161223/a227a02c/attachment.html>