Originally posted by: SystemAdmin

[Hendrik said:]

Hi everybody,

how can I force CPlex to always find the same solutions for the same problem?
Even if I set MIPSearch to 1, I get different results every time I start my program.

Any ideas..?

Best regards,

Hendrik.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[florian_dynasys said:]

Hi.

Can you give some more context? Are you using the Cplex 11 parallel mode? If it's true, have a look to the ParallelMode parameter (opportunistic = not deterministic).

Values

-1  CPX_PARALLEL_OPPORTUNISTIC  Opportunistic  Enable opportunistic parallel search mode 
0  CPX_PARALLEL_AUTO  AutoParallel  Automatic: let CPLEX decide whether to invoke deterministic or opportunistic search, depending on the threads parameter; default 
1  CPX_PARALLEL_DETERMINISTIC  Deterministic  Enable deterministic parallel search mode 


I also think to a time limit.

Florian
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[Hendrik said:]

Hi Florian,

thank you for your answer!
No, I'm not using a parallel version, it's just "plain" CPLEX 11. Here is the output at the beginning of the optimization:
MIP search method: traditional branch-and-cut.
Parallel mode: none, using 1 thread.

Does that answer you question or do you need some other information?

Cheers,

Hendrik.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[florian_dynasys said:]

Well, it's strange.

Are you using a Time Limit? Maybe your CPU use is different between the execution, so the B&C reach different solutions in the given time.

If you have no Time Limit, I don't see.

Florian
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[arunjotshi said:]

Also, please check whether you have set up any value for  MIPGAP.
I would think that if you are asking to CPLEX to stop due to time limit or MIPGAP limit, you might not get the same results every time.
Thanks,
Arun
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[Hendrik said:]

Thanks for your answers.
I'm still investigating, but it seems like the reason is my internal way of building the model. Due to some reason, the order of variables, constraints etc. varies between different runs. Maybe there is some randomization in Java HashMap or some other data structures I use? I don't know. Anyways, if I export the model I can clearly see that the order changes, so that seems to make the difference. If I take the same exported model and import it and solve it, it gives exactly the same solving behaviour again and again, without even setting any extra parameters.
The only thing I wonder is why I haven't observed that before. Could it be that CPlex 9.1 (which I had before) was not affected by that?

Best regards,

Hendrik.
#CPLEXOptimizers
#DecisionOptimization