[Clp] CLP not respecting some constraints with or-tools interface

[Thierry Deo]

Hi,

I've been trying to use or-tools interface with CLP to solve assignment
problems for VRP, and I found out that the solver does not respect all
constraints when I use non unitary demands. Unitary demands work perfectly
fine.

To be more specific :

I choose nb_customers and nb_warehouses.
I generate randomly those customers and warehouses within [0,1000]², and
demands in [1,10] (all integers), and capacities that make sure the problem
is feasible. I then use or-tools MPSolver interface for CLP with binary
variables to indicate if customer i is assigned to warehouse j, and
capacity balancing constraint between depots, and in the solution returned,
there are some customers that weren't assigned to any warehouse

I work with nb_customers=2000, I'm pretty sure I'm ok with my constraint,
since if I use only unitary demands, all 2000 customers are affected, but
when I use random demand between 1 and 10, the problem is said to be solved
to optimality, but only between 1997 and 1999 customers are affected in
most cases. If we use demands in [1,100] , it gets even worse with only
1800 customers affected.

Have someone noticed something like this ?

I can send more details and source code if you like,

Thanks in advance,

Thierry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20130627/66d93a1e/attachment.html>