On Sat, Jul 03, 2004 at 04:58:53PM -0500, Jack L. Stone wrote: > > The point about less effective is the question of "weight". As you know, > the spammer changes the form of attack constantly in an effort to get > around our defenses. Thus, if your "basket" is full of apples and now there > are more "oranges" being used than apples, the defeneses are weaker.
All the learning systems I've seen write-ups for, work best in learning by exception. SA's bayes does better, maybe because it has more bits per token, but even it gets fatigued in many people's experience. My experience isn't that it gets fatigued, but that it gets reinforced and hard to train. Nick
