On Wed, Jul 26, 2006 at 11:56:18AM -0400, Jim Maul wrote: > Hurts more than it helps? Probably not. But it *does* cause weird > things like BAYES_80 being scored higher than BAYES_95. > > body Bayesian spam probability is 80 to 95% BAYES_80 0 0 3.608 2.0 > body Bayesian spam probability is 95 to 99% BAYES_95 0 0 3.514 3.0
Well, it's weird to you because you see a connection between the Bayes rules and probability. The perceptron does not. It simply sees different rules with different hit rates, and scores them based essentially on statistics. http://wiki.apache.org/spamassassin/HowScoresAreAssigned talks about this. -- Randomly Generated Tagline: "Of course my password is the same as my pet's name. My cat's name was Q47pY!3, but I change it every 90 days." - Roddy Vagg
pgp0ZXoQE2BG6.pgp
Description: PGP signature
