[Cbc] cbc infeasible, cplex not
John Forrest
john.forrest at fastercoin.com
Fri May 18 05:52:29 EDT 2018
James,
I have checked in changes in trunk. As there is a small overhead
cleaning up SOS variables, the changes are on an option. See
cleanBounds in CbcModel.hpp for more information.
Preprocessing works for both examples. Keep me informed on problems.
John
On 14/05/18 00:57, James Howey wrote:
>
> Attached lp comes up infeasible in cbc, but finds optimal objective in
> cplex. In cbc I ran it with and without preprocessing which has given
> me problems before.
>
> I ran latest trunk build and an older one. I saw one checkin in latest
> saying SOS shouldn’t have tolerance, and wonder if this is another
> similar manifestation.
>
> It looks like it is correctly computing l1 (63999/64000), but then
> rounding l2 (1/64000 = 1.5625e-5) down to zero. It would seem that a
> consistent rounding strategy would yield a feasible solution. If it
> doesn’t admit l2 = 1.5625e-5 to the solution, it shouldn’t let l4 =
> 1.5930576e-09 in . If it rounds l2 = 1.5625e-5 to zero, it should
> round l1 = .99998089 to 1.0. I would prefer no tolerance.
>
> X0009905: S2:: HS_ST(0)_l1:1 HS_ST(0)_l2:2 HS_ST(0)_l3:3 HS_ST(0)_l4:4
>
> Integer infeasible - objective value 0.00000000
>
> 10 HS_ST(0)_l1 0.99998089 0
>
> 11 HS_ST(0)_l2 0 0
>
> 12 HS_ST(0)_l3 1.9106687e-05 0
>
> 13 HS_ST(0)_l4 1.5930576e-09 0
>
> In the meantime, can anyone suggest a workaround?
>
> Thanks,
>
> jkh
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> https://list.coin-or.org/mailman/listinfo/cbc
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20180518/7ba43bc8/attachment.html>
More information about the Cbc
mailing list