[gecode-users] Gecode 3.0.0: What we will do. What do you want?
Filip Konvička
filip.konvicka at logis.cz
Thu Sep 11 16:43:02 CEST 2008
Hi,
> In more detail, Gecode 3.0.0 will feature:
> - a complete rework of branching, including randomization and support for
> tie-breaking (already finished)
> - abstractions for cost-based optimization (just define a cost function,
> done)
> - a first shot at parallel search (working really great, but integration
> and legal aspects are not yet resolved)
Well actually there's a couple of things :-)
Recently we've been dealing with some constraint systems using Gecode
and we tried to tweak our heuristics, and while doing that, I found out
that I still remember seeing things like backjumping when I first met
CP. So when you mention tie-breaking, randomization - is this somehow
related to that? And (if not) do you think that it's possible to define
some abstractions that would help re-shape the search tree dynamically?
I'm thinking about some "agents" - like actors, but living in the search
engine - that would try to identify the conflicting decisions in the
search tree...
Another thing is a rather technical aspect of parallel search. I was
thinking of using external solvers for efficient consistency-checking of
some special constraint systems. I don't want to go in too much detail
now, the thing is that it would be great if the space created for a new
thread by the parallel-search engine received some sort of thread ID (an
int identifying the thread in the pool would be OK) during the cloning
operation.
> The later versions of Gecode 3.* will gradually improve the search
> capabilities, including a better LDS, abstractions for large neighborhood
> search, automatic configuration of recomputation, etc...
Well then maybe my wishes fit here...
> The long term vision is that Gecode 4.0.0 will be something that is fully
> backed up with high-quality documentation, has very advanced parallel search
> facilities, and might have unique inference modules (elimination of
> redundant propagators, automatic choice of consistency level, SAC, ...).
...and maybe here as well :-)
Cheers,
Filip
More information about the gecode-users
mailing list