stuntman-mike(a)web.de wrote:
Hello,
I noticed that the Leaps algorithm implementation is not existent any more in version
4.x.
I would like to know the reason for discontinuing Leaps support.
Were there performance or implementation problems?
The developer who was working on it lost interest. With no one to
maintain it, we discontinued it. While manners performaned well, it had
performance issues for waltz.
Background: I'm working on my Master's Thesis about pattern
matching algorithms for rule based multiagent systems.
Main motivation is to find an efficient algorithm that needs less space than Rete.
Because Leaps claims to be far more efficient than Rete it is very interessting for me
why you kept Rete and discarded Leaps.
There are things you can do. remove the agenda, but make sure all node
memories are iterated in lifo order - thus propagations will happen in
approximate lifo, but fire straight away. Make sure the network building
orders according to salience. You can then do an adapter algorithm that
chooses to store or not store partial matches, based on demands of the
problem. It isn't hard to write a network that can lazily pull all the
joins from parent nodes (ala leaps) if there are no partial matches -
thus creating a hybrid algorithm.
Thank you in advance.
Regards, E.L.
_____________________________________________________________________
Der WEB.DE SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!
http://smartsurfer.web.de/?mc=100071&distributionid=000000000066
_______________________________________________
rules-dev mailing list
rules-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev