On Thu, 20 Aug 2009 20:58:02 -0500 Chris <cpoll...@embarqmail.com> wrote:
> Ran it through sa-learn as spam, spamassassn -r and -t which scored > much better: > ... > > 5.0 BAYES_99 BODY: Bayesian spam probability is 99 to Unfortunately that's not a good guide to future performance. Most spams contain a lot of unique tokens that skew the result if you retest the same spam after learning.
