[ODE] Mismatching results / non-deterministic behaviour
D.Hauser at dion-software.com
Thu Sep 14 07:25:46 MST 2006
Ah, I see!
That explains a lot of things I've experienced with repeating steps
(or trying to at least) actually! ;)
That also explains, why results were the same, when being run on
different machines, because the ODE random number was never
initialised and therefor using the same predefined seed.
> -----Ursprüngliche Nachricht-----
> Von: ode-bounces at q12.org [mailto:ode-bounces at q12.org] Im
> Auftrag von Klaus Backert
> Gesendet: Donnerstag, 14. September 2006 15:43
> An: ode at q12.org
> Betreff: Re: [ODE] Mismatching results / non-deterministic behaviour
> Am 14.09.2006 um 09:13 schrieb Hauser Dietmar:
> > Can somebody confirm this?
> > I've skimmed over the code quickly and there are quite a few rands
> > in there.
> > However, so far, the ODE seemed very deterministic to me...
> > --
> > dIdl
> In the Quickstep algorithm there is
> #define RANDOMLY_REORDER_CONSTRAINTS 1
> and, depending on that, a piece of code for randomly shuffling the
> constraints in the LCP solver for the contact and friction model.
> Similar things are going on in the Stepfast algorithm with
> #define RANDOM_JOINT_ORDER
> (but Stepfast has been superseded by Quickstep).
> I don't know exactly, but it looks like there is some random
> influence when stepping the ODE world with Quickstep. And if that is
> true, using Quickstep, results are reproducible only if the random
> seed is reset, as Chris mentioned already.
> ODE mailing list
> ODE at q12.org
More information about the ODE