Actually, I think I mean "by contract". Consider having a set rules and
facts that would run in both CLIPS and JESS. Wouldn't you expect to get the
same results, independent of which inference engine you used?
There are some things that are implementation dependent, for example which
rule is fired next when multiple rule activations have the same precedence.
But since this is a non-procedural environment, it should not matter which
one is actually fired, by definition. The end results should be equivalent,
though they may be reached by different paths. If it matters then you have a
procedural, order dependent rule set.
I strongly agree that any dependence on implementation peculiarities should
be avoided. One test of this which I use is to randomly rearrange the order
of my rules and facts in the .clp files. The rearrangement should not effect
the correctness of the results.
- Lewis
-----Original Message-----
From: Thomas Barnekow [mailto:[EMAIL PROTECTED]]
Sent: Wednesday, January 03, 2001 3:02 PM
To: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: RE: JESS: Triggering of Rules
>Hi!
>
>Lewis Hart wrote:
>> I have to disagree, the triggering of rules in JESS is completely
>> deterministic. Given the same initial conditions and the same input facts
>> asserted in the same order, -- the same rules in the same order will
always
>> be fired. This even includes random number generators if you can control
>> their initial "seed" values. Knowledge/rule-based systems are not
>> procedural, and often the order of rule firing is not obvious beforehand,
>> but this is different issue.
>
>Chris Glur wrote:
>> * AI systems are often non-deterministic, like sorting potatos sizes
>> through 3 sieves - you can't be sure which hole a particular potato
>> will pass, to get the correct selection.
>
>You have to differentiate between determinism "by contract" (I assume Chris
means this one) and "by >implementation" (that's probably what Lewis is
talking about.) That is, a system can be non-deterministic >by contract (or
conceptually) while its actual behaviour, as implemented, can be "somehow
>deterministic." Be careful, though, to rely upon this "accidental"
determinism, because the behavior >might change if any modifications to the
implementation are made.
>
>Thomas
---------------------------------------------------------------------
To unsubscribe, send the words 'unsubscribe jess-users [EMAIL PROTECTED]'
in the BODY of a message to [EMAIL PROTECTED], NOT to the
list (use your own address!) List problems? Notify [EMAIL PROTECTED]
---------------------------------------------------------------------
