Re: Problem with fuzzy rule implication

WSiler (
Sun, 9 Nov 1997 04:54:50 +0100 (MET)

> In an earlier version of our software we fired BOTH the then and else
> statements with different truths. This turned out to be a knowledge
> engineering nightmare.

I'm in agreement with Earl's method of handling the if-then-else rules. Our
attempt to fire both consequents when both the antecedent cofidence and (1 -
antecedent confidence) are above the rule-firing threshold had exactly the
results that Earl describes - a hopeless mess for other than toy problems.

About the only time I've wished we had the if-then-else construct is when the
rule antecedent has only crisp attribute values. When the antecedent
attributes are fuzzy, I'm happier with the single consequent.

Bill Siler