[Cbc] Cost function item order

[John Forrest]

Yves,

I will track it down.  Somehow (maybe correctly) an element of about 
1.0e-7 is being created by preprocessing.  In most cases the LPs are 
fine, but somehow in this case one LP is infeasible by 1.2e-7.

As a temporary fix, I would switch off scaling and set the primal 
tolerance to 5.0e-7.  Either of these fixes problem on your example - 
the second one should be all you need, but I don't think it will do any 
harm to switch off scaling.

I will get back to you when I have more to report.

John Forrest
On 08/01/14 14:55, Yves Touchard wrote:
> John,
>
> I have the same kind of problem than (Vol78, Issue 6): swapping two 
> lines of the cost function to minimize modifies the solver result 
> (problem infeasible or not).
>
> What i did:
>
>   * Download the Cbc-2.8.8 archive
>     (http://www.coin-or.org/download/source/Cbc/Cbc-2.8.8.tgz)
>   * Inflate it (tar xzf Cbc-2.8.8.tgz)
>   * Copy the Vol78,Issue 6 fix you recently did: cp
>     https://projects.coin-or.org/Cgl/browser/stable/0.58/Cgl/src/CglProbing/CglProbing.cpp
>     Cbc-2.8.8/Cgl/src/CglProbing/CglProbing.cpp
>   * Build the binaries: ./configure -C; make; make test; make install
>   * Solve the attached problem: Cbc-2.8.8/bin/cbc test.lp -> The
>     problem is declared as infeasible
>   * Swap lines 112 & 113 (those surrounded by blank lines) and solve
>     again -> The optimal solution is found.
>
> Obviously, i could remove these lines that are not used by the cost 
> function since mul coefficients are null. But, this constraint file is 
> generated by a piece of code and it is easier to keep these 
> unnecessary lines in the cost function (i expect the solver ignore them).
>
> Any help will be greatly appreciated.
>
> Regards,
>
> Yves
>
>
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20140109/34e0ac9f/attachment.html>