[Coin-discuss] How to speed-up Cbc?

[John J Forrest]

This is a case of less is more - using trunk

cbc -solve                                              9.81
cbc -cuts off -solve                                    1.08
cbc -cuts off -preprocess off -solve                    0.71
cbc -cuts off -strong 0 -preprocess off -solve          0.53
cbc -cuts off -strong 0 -preprocess off -heuristic off -solve   0.30

As you can see the cuts are the main culprit.  The code thinks they are 
doing good but they do not seem to do much.  Interestingly neither flow 
cover or mixed integer rounding give any cuts - while they do in cplex.

For larger problems it may still be worth doing preprocessing.  Also 
switching back on feasibility pump is marginally better.

(Note - I have just updated trunk so that strong=0 switches off strong 
branching altogether - with previous version it would be -strong 0 -trust 
0)

John Forrest





Michal Kaut <mail at michalkaut.net> 
Sent by: coin-discuss-bounces at list.coin-or.org
01/26/2007 08:06 AM
Please respond to
michal.kaut at hiMolde.no; Please respond to
Discussions about open source software for Operations Research 
<coin-discuss at list.coin-or.org>


To
coin-discuss at list.coin-or.org
cc

Subject
[Coin-discuss] How to speed-up Cbc?






Hello,

I am working with a MIP problem. I have tried a couple of solvers and 
Cbc is by far the slowest one. Can someone give me a hint which options 
can I use to speed this particular problem up? I attach an MPS file.

On my computer, CPLEX uses under 0.5 seconds, glpsol (from GLPK) 0.2 
seconds(!) and lpsolve 3.5 seconds, while Cbc with default settings 
almost 16 seconds. I tried both my own compiled binary (Windows with 
MSYS and MinGW) and both binaries from the "Coin Binary" project.

I tried several options and found that I can get the time down to 6.5 
seconds using "-heur off", but it is still much slower than the rest.
Note that this just a test example. I have a bigger version that takes 
under 40 seconds with both CPLEX and glpsol, while Cbc did not solve it 
in more than 20 minutes (then I gave up and stopped it).


Any help would be appreciated.

Regards,
Michal Kaut
[attachment "small_test.zip" deleted by John J Forrest/Watson/IBM] 
_______________________________________________
Coin-discuss mailing list
Coin-discuss at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/coin-discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/coin-discuss/attachments/20070126/aa2735ba/attachment.html>