[Cbc] Performance issues

[John Forrest]

Berhard,

I agree with Haroldo on some points, but with an event handler and 
feasibility pump I get a solution time of 0.044 seconds - see attached.

Will look to see how to improve build time.

John Forrest
On 10/01/17 15:26, Haroldo Gambini Santos wrote:
> Hi Bernhard,
>
> Did some experiments with your MIP.
>
> With default options CBC really performs poorly, but changing a few 
> options it solves it almost instantly.
>
> try options:
> cbc test.lp zero on heur off reduce ifmove ratio 0.005 latwomir 
> endonly lagomory endonly clique off knapsack off threads 4 combine2 
> on  solve
>
> It is solved (very close to optimality) in 2 seconds in my core i7.
>
> What I noticed is that your model is "integer friendly": most 
> heuristics can be turned off and it still finds a good solution. So 
> expensive heuristics such as feasibility pump (which are mostly useful 
> in problems where finding any feasible solution is hard) can be turned 
> off saving some time.
>
> Cheers,
>
> Haroldo
>
>
>
> Em 10/01/2017 12:08, Bernhard Dick escreveu:
>>> I got the message
>>>
>>>
>>> Error Could not decrypt data (Wrong key ?)
>>>
>>> While following the link. It seems that it is encrypted.
>> Hm seems that the URL got broken. I've attached the lp file now. (I 
>> didn't
>> want to post  it onto the mailing list as it's almost 100kB in size)
>>
>>    Regards
>>      Bernhard
>>>
>>>
>>>
>>>
>>> Em 10/01/2017 11:47, Bernhard Dick escreveu:
>>>
>>>
>>>     Hi Haroldo,
>>>
>>>
>>>         Could you send us the LP file so that we can play with some
>>> parameters ?
>>>
>>>     Even more, I'm able to send you the C++ source. I only had to
>>> obfuscate the
>>>     variable names. The source is attached to this mail, the generated
>> lp
>>> file
>>>     can be found here:
>>>     https://bin.jjim.de/paste/T2wGOAHI#tvUwPsvzhYejnk57dTi++Buf9J
>>> 3heraZRzDTbWX89
>>>     7P
>>>
>>>       Regards
>>>         Bernhard
>>>
>>>
>>>         Cheers,
>>>         Haroldo
>>>
>>>         Em 09/01/2017 06:21, Bernhard Dick escreveu:
>>>
>>>             Hi,
>>>
>>>             I have a MILP with ca 2000 variables, 400 of them
>>> being binary and
>>>             1300 constraints. Currently I solve this using
>> lpsolve
>>> but I also
>>>             tried it using CBC, based on the sample5.cpp with
>>> CoinBuild. However I
>>>             face some performance issues: On my test system
>>> lpsolve needs about
>>>             120msec to solve the problem, CBC needs ca 3-4secs
>>> is thus 30-40 times
>>>             slower. I use a MID-GAP of 3% set by
>>>
>>>         model.setAllowablePercentageGap(3).
>>>
>>>             Another interesting fact is that If I save the
>> problem
>>> to an lp oder
>>>             mps file and reload the file in CBC solving gets
>> about
>>> 1sec faster.
>>>             Do you have any ideas how to improve the
>>> performance here, as in
>>>             different benchmarks that I've found cbc performs
>>> much faster than
>>>             lpsolve, so I'm a bit surprised by the results I
>> see.
>>>                Regards
>>>                  Bernhard Dick
>>>
>>>
>>>     _______________________________________________
>>>             Cbc mailing list
>>>             Cbc at list.coin-or.org <mailto:Cbc at list.coin-or.org>
>>>             https://urldefense.proofpoint.com/v2/url?u=http-
>>> 3A__list.coin-2Dor.org
>>>             _mailman_listinfo_cbc&d=CwICAg&c=Ngd-
>>>
>>>         ta5yRYsqeUsEDgxhcqsYYY1Xs5ogLxWPA
>>>
>>>             _2Wlc4&r=pLOfVNEEHf-xhIqn1-
>>>
>>>         uzYcZ6Q7UefG6Bg6rXCKTMiAA&m=IFlfsGIm5yclwUs
>>>
>>>             Liun-5eSbIc-LztVKfjJfE-
>>>
>>>         w0iqk&s=j4C13Bzc90fB9iNOGEw28UDXreXgbHI7seB7n5x
>>>
>>>             u7r0&e=
>>>
>>>
>>>         --
>>>
>>>     ==================================================
>>>         Haroldo Gambini Santos
>>>         D.Sc, Computer Science
>>>         Universidade Federal de Ouro Preto
>>>         http://www.decom.ufop.br/haroldo/
>>>
>>>
>>>
>>>
>>> -- 
>>> ==================================================
>>> Haroldo Gambini Santos
>>> D.Sc, Computer Science
>>> Universidade Federal de Ouro Preto
>>> http://www.decom.ufop.br/haroldo/
>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: solver.log
Type: text/x-log
Size: 3323 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20170110/df57485c/attachment-0002.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: solver-mod2.cpp
Type: text/x-c++src
Size: 11808 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20170110/df57485c/attachment-0003.bin>