Re: [Choco-users] Choco Speed

[Maurice D. <Mau...@en...>]

Le 10 d=E9c. 04, =E0 13:06, Mark Utting a =E9crit :

> I've used the N queens problem (ALL solutions -- details below) to
> benchmark various integer/finite-domain solvers to compare their basic
> efficiency of triggering constraints, labeling and backtracking.

The problem with looking for all solutions is that you are testing
the performance of "propagation engine" + "enumeration solver".
My feeling is that the enumeration solver from choco could be
improve and it is not the main advantage of choco.

In the other side, it would be nice to have such a test for the
"propagation engine" only, which is the core and the main advantage
of choco.

Also your problem, which is the same as a previous one of mine,
is that choco seems too be design for somebody which eventually use
other tools as a brick, and not the opposite as you'like to.


So I'd hope a stong separation between these two fonctionnality of
Choco :
- the "propagation engine"
- the enumeration system which all the "Solver" "Solve" "Solving"
   "Solvare" and other "BranchingTrace" classes

I hope to provide a "solver lite" system, easiest to personalised,
while keeping compatibility with choco current features.
But this projet is now sleeping as I'm belated on other projects.

Anyhow, thank your for your bench, and I hope some choco guru
will had and maintain such a bench with the choco distribution.


> 3. Is Choco.sf.net likely to be a good option for us, if we want to =
use
>   just its integer solver, but extend it with hooks to report changes
>   in domains etc?

It depend of what you call "Solver".
You could build you own "dedicate enumSolver lite",
and only use th constraint propagation system from
choco.

-- Maurice Diamantini