<div>
Hi Michael,
</div>
<div><div><br></div></div><blockquote type="cite" style="border-left-style:solid;border-width:1px;margin-left:0px;padding-left:10px;"><span><div>You experienced a speed up factor of 2; on which test set or test</div><div>instances, are these instances similar in their structure?</div></span></blockquote><div>In all my tests I used sets of similar problems generated by AMPL. E.g. a set of ten traveling salesman problem stubs generated from the same model file but with different random seed.</div><div><br></div><blockquote type="cite" style="border-left-style:solid;border-width:1px;margin-left:0px;padding-left:10px;"><span><div>We are sure that parameter for a particular instance type can be improved,</div><div>but how does your tuning works on an instances set like the MIPLIB?</div></span></blockquote><div>Improving parameters for a particular instance type was the main use case for the tool when I was working on it. I think right now it is not very useful for the full MIPLIB at least because there are problems which hit limits with SCIP.</div><div><br></div><blockquote type="cite" style="border-left-style:solid;border-width:1px;margin-left:0px;padding-left:10px;"><span><div>Do you check the different results of the runs for feasibility/optimality</div><div>or compare the solution value for each instance?</div></span></blockquote><div>Yes, I compare the SCIP status and the primal value to the values from runs with default parameters.</div><div><br></div><blockquote type="cite" style="border-left-style:solid;border-width:1px;margin-left:0px;padding-left:10px;"><span><div>Is Turbo Boost and Hyper-threading disabled for your results and are the</div><div>runs performed exclusively on a computer?</div></span></blockquote><div><div><div>I run SCIP on virtual machines running under KVM hypervisor in a commercial cloud so I have no such control. The performance of a VM usually depends on the load made by other customers on the same hypervisor host. That is why I try to run SCIP with default parameters for every problem on every node. I can compare this time to other times I get on the same host.</div><div><br></div><div><br></div><div>Best regards,</div><div>Sergey</div></div><blockquote type="cite"></blockquote></div><div><br></div><div> </div><blockquote type="cite" style="border-left-style:solid;border-width:1px;margin-left:0px;padding-left:10px;"><span><div><br></div><div>Best, Michael</div><div><br></div><blockquote type="cite"><div><div>Hi Sergey,</div><div><br></div><div>I would be interested not so much in the implementation, but in the</div><div>methodological background.</div><div><br></div><div>Firstly it estimates how long it takes to solve single problems with</div><blockquote type="cite"><div>default parameters.</div></blockquote><div><br></div><div><br></div><div>How would you do that, in particular for those runs that hit the time</div><div>limit? I believe a LOT of people would be interested in such an estimate.</div><div><br></div><div><br></div><div><br></div><blockquote type="cite"><div><div>Then we do a single run (for every problem) for every special value</div><div>(e.g.</div><div>-1 or 0) of every parameter and two runs for ranged parameters (bigger</div><div>and</div><div>lower than default values). That is about 3000 runs for every problem in</div><div>a</div><div>test set so I use a cluster of virtual machines in a cloud for the</div><div>computations. After this step the tool compares the impact which</div><div>parameters</div><div>and their values make on the solving time.</div></div></blockquote><div><br></div><div><br></div><div>How? Considering a single parameter at a time? Or their combination? Or</div><div>pairs? Or ...</div><div><br></div><div><br></div><blockquote type="cite"><div><div>Finally the tool takes four parameters (and their values) with the</div><div>biggest</div><div>impact and finds their best combination. This step is repeated for a fey</div><div>times</div></div></blockquote><div><br></div><div><br></div><div>"This step"? Why repeating the four-parameter-combinations? For taking</div><div>into</div><div>account performance variability?</div><div><br></div><div><br></div><blockquote type="cite"><div><div>and we get a settings file which speeds up computations noticeably for</div><div>the</div><div>given set of problems.</div></div></blockquote><div><br></div><div><br></div><div>Cool. Still would be interested in details.</div><div><br></div><div><br></div><div>Take care</div><div>Marco</div><div><br></div><div><br></div><div><br></div><div><br></div><div>--</div><div>Prof. Dr. Marco Lübbecke</div><div>RWTH Aachen University</div><div>Chair of Operations Research</div><div>Kackertstrasse 7</div><div>D-52072 Aachen</div><div>Germany</div><div><br></div><div>fon / fax: +49 241 80-93362 / 92369</div><div><a href="mailto:marco.luebbecke@rwth-aachen.de">marco.luebbecke@rwth-aachen.de</a></div><div><a href="http://www.or.rwth-aachen.de/luebbecke">www.or.rwth-aachen.de/luebbecke</a></div><div><br></div><div><<a href="http://scholar.google.de/citations?user=cYIpR60AAAAJ">http://scholar.google.de/citations?user=cYIpR60AAAAJ</a>></div><div><<a href="https://www.researchgate.net/profile/Marco_Luebbecke/">https://www.researchgate.net/profile/Marco_Luebbecke/</a>></div><div><<a href="http://www.informatik.uni-trier.de/%7Eley/pers/hd/l/L=uuml=bbecke:Marco_E=.html">http://www.informatik.uni-trier.de/%7Eley/pers/hd/l/L=uuml=bbecke:Marco_E=.html</a>></div><div><<a href="https://www.xing.com/profiles/Marco_Luebbecke">https://www.xing.com/profiles/Marco_Luebbecke</a>></div><div><<a href="https://plus.google.com/u/0/100665281566037430285/">https://plus.google.com/u/0/100665281566037430285/</a>></div><div><<a href="http://de.linkedin.com/pub/marco-l%C3%BCbbecke/7b/a2a/77/">http://de.linkedin.com/pub/marco-l%C3%BCbbecke/7b/a2a/77/</a>></div><div><<a href="https://www.facebook.com/OperationsResearchRWTH">https://www.facebook.com/OperationsResearchRWTH</a>></div><div><<a href="https://twitter.com/#%21/mluebbecke">https://twitter.com/#%21/mluebbecke</a>> <<a href="http://klout.com/#/mluebbecke">http://klout.com/#/mluebbecke</a>></div><div>_______________________________________________</div><div>Scip mailing list</div><div><a href="mailto:Scip@zib.de">Scip@zib.de</a></div><div><a href="http://listserv.zib.de/mailman/listinfo/scip">http://listserv.zib.de/mailman/listinfo/scip</a></div></div></blockquote></span>
</blockquote>
<div>
<br>
</div>