glebbelov
commented
4 years ago Hi,
you probably mean different optimal objective value, not just solution.
Depending on the MZN version, Gurobi might receive a different FlatZinc representation of your instance. In the current develop branch, it's quadratics which are given to Gurobi vs decomposed for other MIP solvers. If you have no multiplications, the FZN must be same. YOu can check if it's the same by adding --writeModel model.mps and running each solver directly from each mps instance.
If they receive effectively the same instance, then it's either a bug, or, much more common and even normal, numerical inaccuracy. In the verbose output of a solver you can see numerical stability characteristics and warnings. If there are, try increasing numerical emphasis, or better, reformulate the model. Otherwise, it could be a bug and you can report it.
Cheers, Gleb
On Sat, 29 Aug 2020 at 19:28, stef-1 notifications@github.com wrote:
When running a MIP with "minizinc --solver cplex -v -p 8 --solver-statistics model.mzn data.dzn", another solution to the specific instance is found than with Gurobi "minizinc --solver gurobi-v -p 8 --solver-statistics model.mzn data.dzn". However, with the additional option "--no-optimize", CPLEX provides the same optimal solution, that Gurobi finds in the first place. This is only a problem for a specific problem instance (1 out of 4200). If needed, I will provide the model and data files on request.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/MiniZinc/libminizinc/issues/405, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEEBOZQL53AND6DJRWQNN73SDDC5NANCNFSM4QO4N4CQ .
-- Dr Gleb Belov Monash University +61 3 9903 1622
Dekker1
When running a MIP with "minizinc --solver cplex -v -p 8 --solver-statistics model.mzn data.dzn", another solution to the specific instance is found than with Gurobi "minizinc --solver gurobi-v -p 8 --solver-statistics model.mzn data.dzn". However, with the additional option "--no-optimize", CPLEX provides the same optimal solution, that Gurobi finds in the first place. This is only a problem for a specific problem instance (1 out of 4200). If needed, I will provide the model and data files on request.