[Couenne] Analyzing output

Pietro Belotti pbelott at clemson.edu
Thu Feb 28 12:12:29 EST 2013


Ina,

> One more question concerning the output "...best solution":
> For some instances Couenne gives out "1e+050 best solution" at the 
> beginning (before an integer solution was found). I guess this is an 
> upper bound estimation before starting branch-and-bound, or is it 
> computed in a different way?

Yes. It simply means no feasible solution is found so the upper bound is 
+infinity.

>>> How do I know how every new solution was found? Does the output give 
>>> information which heuristic found the solution?
>>
>> Yes, every line containing "integer solution". However, only those that 
>> improve on the previous value of the cutoff.
>
> I don't see the information about how the solution was found. If the 
> output is "Integer solution of 1166.8 found after 450865 iterations and 
> 21806 nodes (587.68 seconds)", does this mean the solution was found 
> without a heuristic just by solving a node in the b&b-tree?

Sorry for overlooking this in my answer. Feasible solutions can be found 
by a heuristic, by the LP solver (when the LP solution happens to be 
feasible for the MINLP as well) and by strong branching, if I remember 
correctly. What you quote is the typical output when the solution is found 
by the LP solver.

> If so, would this line give the information about the heuristic that 
> found the solution, if the was found by a heuristic?

Yes.

> One more question: Is it possible to get information about the memory 
> used during the solving process?

Couenne does not keep track of it, and I'm not sure if Cbc does. Maybe 
you can use some generic command line tools, but I am not knowledgeable 
enough to help you on this.

Pietro

--
Pietro Belotti
Dept. of Mathematical Sciences
Clemson University
email: pbelott at clemson.edu
phone: 864-656-6765
web:   http://myweb.clemson.edu/~pbelott

On Thu, 28 Feb 2013, Liffey1986 at gmx.de wrote:

> Thanks Pietro, your mail was very helpfull!
>
> One more question concerning the output "...best solution":
> For some instances Couenne gives out "1e+050 best solution" at the 
> beginning (before an integer solution was found). I guess this is an 
> upper bound estimation before starting branch-and-bound, or is it 
> computed in a different way?
>
>>> How do I know how every new solution was found? Does the output give 
>>> information which heuristic found the solution?
>>
>> Yes, every line containing "integer solution". However, only those that 
>> improve on the previous value of the cutoff.
>
> I don't see the information about how the solution was found. If the 
> output is "Integer solution of 1166.8 found after 450865 iterations and 
> 21806 nodes (587.68 seconds)", does this mean the solution was found 
> without a heuristic just by solving a node in the b&b-tree? If so, would 
> this line give the information about the heuristic that found the 
> solution, if the was found by a heuristic?
>
> One more question: Is it possible to get information about the memory 
> used during the solving process?
>
> Best,
> Ina
>
>
>
>
> -------- Original-Nachricht --------
>> Datum: Thu, 28 Feb 2013 11:41:35 -0500 (EST)
>> Von: Pietro Belotti <pbelott at clemson.edu>
>> An: Liffey1986 at gmx.de
>> CC: Couenne mailing list <couenne at list.coin-or.org>
>> Betreff: Re: [Couenne] Analyzing output
>
>> Ina,
>>
>>> How do I know when the first solution is found? Is this when the output
>>> gives for the first time "integer solution of .... found" or when it
>>> says " .... best solution"?
>>
>> The line "integer solution..." indicates that one solution was found and
>> it is stored and used as a cutoff. This happens before the output line
>> containing "... best solution".
>>
>>> How do I know how every new solution was found? Does the output give
>>> information which heuristic found the solution?
>>
>> Yes, every line containing "integer solution". However, only those that
>> improve on the previous value of the cutoff.
>>
>>> If Couenne prints some additional statistics I get an extra line at the
>>> end of the output. What do the following brackets mean:
>>
>> [root] the number of linear inequalities that provide a relaxation of the
>> nonlinear problem that added at the root node;
>> [tot]  the total number of such linear inequalities added;
>> [sep]  the time spent in the separation of these inequalities;
>> [bb]   the time spent within the branch-and-bound
>>
>> This is printed out with the option "display_stats yes" in the couenne.opt
>> file, although it is used mostly for debugging purposes.
>>
>> I hope this helps.
>>
>> Regards,
>> Pietro
>>
>> --
>> Pietro Belotti
>> Dept. of Mathematical Sciences
>> Clemson University
>> email: pbelott at clemson.edu
>> phone: 864-656-6765
>> web:   http://myweb.clemson.edu/~pbelott
>>
>> On Thu, 28 Feb 2013, Liffey1986 at gmx.de wrote:
>>
>>> Hi all,
>>>
>>> I'm trying to analyze the Couenne output, but there is still a few
>> things I'm not sure about:
>>>
>>> How do I know when the first solution is found? Is this when the output
>> gives for the first time "integer solution of .... found" or when it says "
>> .... best solution"?
>>>
>>> How do I know how every new solution was found? Does the output give
>> information which heuristic found the solution?
>>>
>>> If Couenne prints some additional statistics I get an extra line at the
>> end of the output. What do the following brackets mean:
>>> [root]
>>> [tot]
>>> [sep]
>>> [bb]
>>>
>>> Best Regards
>>> Ina
>>> _______________________________________________
>>> Couenne mailing list
>>> Couenne at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/couenne
>>>
>


More information about the Couenne mailing list