[Cbc] cpx vs cbc: different objective for milp

[Matthew Galati]

An update... Xpress using the .mps file (associated with this .lp) also
gives 7.113e4.

However, when I run this in memory (i.e., I construct the model from raw
data in an OSI object) and solve using OsiCBC, I get 7.0413e4 AND if I use
OsiCPX, I also get 7.0413e4. So.... now I am thinking the real optimal is
probably 7.0413e4 and the 7.11e34 is related to some kind of trunctation
with .mps or .lp. Thoughts?

This does not explain why CBC and CPX would give different solutions against
the same input file (.lp).  Perhaps the two read in using a different
precision? I would expect both solvers are using doubles and read in as many
digits as are given - but I am not sure.




CPX10.2 gives optimal solution of 7.1113e4
> CBC2.3 gives optimal solution of  7.0413e4
>
> for http://coral.ie.lehigh.edu/~magh/tmp/cbc_vs_cpx.lp<http://coral.ie.lehigh.edu/%7Emagh/tmp/cbc_vs_cpx.lp>
>
> I am not entirely sure, but I am guessing CBC is wrong. Let me know if this
> can be fixed.
>
> Thanks,
> Matt
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.coin-or.org/pipermail/cbc/attachments/20090730/444580d5/attachment.html