Difference in solvers output

[ursula challita]

Hi all,

I am working on comparing two solvers (gurobi and cplex).
I built an LTE network planning problem in AIMMS, however, sometimes
the selected base stations are different (knowing that the objective
is the same). Why is that??
I would really appreciate any help.

Thanks in advance,
Ursula

[Luis Pinto]

Hello Ursula,

If I understand correctly, this means that there are multiple solutions with the same "optimal" result.

From what I gather, CPLEX and GUROBI have a deterministic "path" to a solution (and therefore offer the same result in two separate runs) given certain conditions:

1 - the initial basis used (if any) is the same in 

2 - You are not using the multi-thread with an opportunistic approach

3 - Your stop condition isn't time based

Of course, what I mean in two separate runs, is CPLEX running twice will offer the same solution and GUROBI will as well, but that does not guarantee that they will offer the same solution among each other.

Hope it helps,

Cheers,

Luis Pinto

www.unisoma.com.br

--
You received this message because you are subscribed to the Google Groups "AIMMS - The Modeling System" group.
To post to this group, send email to ai...@googlegroups.com.
To unsubscribe from this group, send email to aimms+un...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/aimms?hl=en.

[ursula challita]

Dear Luis,

Basically I want to know why the outcome was different.
All the initial parameters are the same.
Shouldn't both solvers lead to the same soltion among each others?

Thanks again,
ursula

On May 14, 4:09 pm, Luis Pinto <luisf...@gmail.com> wrote:
> Hello Ursula,
>
> If I understand correctly, this means that there are multiple solutions
> with the same "optimal" result.
>
> From what I gather, CPLEX and GUROBI have a deterministic "path" to a
> solution (and therefore offer the same result in two separate runs) given
> certain conditions:
> 1 - the initial basis used (if any) is the same in
> 2 - You are not using the multi-thread with an opportunistic approach
> 3 - Your stop condition isn't time based
>
> Of course, what I mean in two separate runs, is CPLEX running twice will
> offer the same solution and GUROBI will as well, but that does not
> guarantee that they will offer the same solution among each other.
>
> Hope it helps,
>
> Cheers,
>
> Luis Pinto
>
> www.unisoma.com.br
>

> On 14 May 2012 04:12, ursula challita <ursulachall...@gmail.com> wrote:
>
>
>
> > Hi all,
>
> > I am working on comparing two solvers (gurobi and cplex).
> > I built an LTE network planning problem in AIMMS, however, sometimes
> > the selected base stations are different (knowing that the objective
> > is the same). Why is that??
> > I would really appreciate any help.
>
> > Thanks in advance,
> > Ursula
>
> > --
> > You received this message because you are subscribed to the Google Groups
> > "AIMMS - The Modeling System" group.
> > To post to this group, send email to ai...@googlegroups.com.
> > To unsubscribe from this group, send email to
> > aimms+un...@googlegroups.com.
> > For more options, visit this group at

> >http://groups.google.com/group/aimms?hl=en.- Hide quoted text -
>
> - Show quoted text -

[Luis Pinto]

Not really.
If there are multiple optimal solution, in other words, more than one solution that generates the best possible objective function, then each solver can come to different results. This is because there is a lot of "intelligence" within a solver in order to find the optimal solution. For instance, in MIP's, how to explore the branch and bound tree.
But even within a LP, since the selection of a variable from one basis to another can be based on "heuristics".

Still, this should only happen if there are more than 1 optimal solution. A degenerate model.

Not sure if this is what you want, but you could perturb your matrix, in order to eliminate degeneracies,  by adding a small coefficient to parameters, although I would not suggest it.

Hope it helps,

Cheers,


Luis Pinto

www.unisoma.com.br