[Cbc] Segfault solving a problem containing SOSs

[Luís Borges de Oliveira]

John,

On 03-11-2016 19:08, John Forrest wrote:
> Fixed the assertion fail.
>
> I don't think, for difficult problems, it is possible to get good 
> results by replacing constraints by branching rules.
>
> I think a better approach is to make the LP solver faster for problems 
> with many constraints.
>
> I have added driverFat.cpp to Cbc/examples.  It is not tidied up, but 
> may improve things for you.  For a previously posted lp - test.lp with 
> 998212 rows and 106780 columns it solved completely in under three 
> hours (okay - using four threads).
>
> One could probably do better if the driver was more intelligent on 
> strong branching - which would not take long to code.  As it was, I 
> just turned that off.

I tried to solve the big test.lp with driverFat (single-threaded) but it 
didn't reach any solution within ~70h. Also, I've found that driverFat 
solves the smaller lp much slower than plain Cbc (even if I disable 
strong branching).

In any case, at this point, I will follow your advice and drop the SOS 
approach. Cbc is working pretty well for smaller problems and I'll 
decompose larger ones.

Thanks for your help,
Luís
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20161110/65193a64/attachment.html>