Hi,
Le 2 mai 09 à 17:37, Tom Lane a écrit :
> My knowledge of AI search algorithms is about 20 years obsolete, but
> last I heard simulated annealing had overtaken genetic algorithms for
> many purposes. It might be interesting to try a rewrite based on SA;
> or maybe there's something better out there now.
I've done very very few courses in the domain, and was tough SA too
(about 10 years ago). But what I gathered more recently was that it's
already being obsoleted by fuzzy logic ideas, which implementations
are more and more reliable.
The idea would be to offer typical queries and possible plans, and to
tell apart the very good ones from the good and pretty bad ones. This
would serve as a model for the fuzzy logic engine to take decisions
and be able to give back best possible plan (as showed) for a given
input set (the query).
I'm very unclear how to better specify learning methods or models,
etc, and probably I should have left this part away. But it well seems
it's the "better out there now" trend.
Recreating the same end-product from a given recipe and new constraint
(raw milk enabled industrial products is forbidden in this country) is
an example of application.
Regards,
--
dim