Deterministic behaviour

[Iain]

Hi,

I'm solving a series of about 1000 MILPs, the result of one MILP
affects the next MILP.
I record the value from each of these MILPs, and at the end I total
all these values (say, total revenue over a day or something like
that.)

Everytime I run it on the same data set, I get different results. I
see some discussion in the reference docs about determinism for
solving an LP, but the MILP parameters page doesn't say anything.

How do I guarantee deterministic beaviour?

Cheers,
Iain

[Greg Glockner]

Gurobi was designed to be deterministic, meaning that it will produce the same results so long as you don't change the computer, Gurobi version, matrix, or parameters. The known exceptions are:
1. If you are setting a time limit
2. With method=3 (concurrent)
3. On some 32-bit Linux systems, you can get different results; this is believed to be caused by the compiler that we use to build Gurobi

If you're using the latest version (4.6.1), you aren't changing anything and you don't fit one of the exceptions above, please post a follow-up so that we can try to understand the issue better. Thanks.

[stefano.gualandi]

Is the case I set a "solution limit" another expection to deterministic behaviour with 8 threads?

thanks,

Stefano

[Greg Glockner]

> Is the case I set a "solution limit" another expection to deterministic behaviour with 8 threads?

Yes.

[Greg Glockner]

Sorry, I misunderstood your question:

>> Is the case I set a "solution limit" another expection to deterministic behaviour with 8 threads?
>
> Yes.

The solver is designed to give deterministic behavior when setting a solution limit. You contacted us privately with an example where this is not happening; we are investigating that.