On Tue, 2003-09-30 at 13:46, Peter Chubb wrote: > The false negatives mostly have MICROSOFT_EXECUTABLE and BAYES_01 > set, which adds to a score of -0.2. I've been saving these to a > file, and running sa-learn on them, along with learning as ham the > rest of the email I get, but it seems to be learning really slowly. > > Is there any good way to seed things up? I'm seeing 40 to 50 SVEN > worm emails an hour; around 10 get through to my INBOX, the rest > are classified correctly.
I'm no expert by any stretch of the imagination but from my personal experiences 2 things come to mind. By default the bayes filter doesn't kick in until 200 odd emails have been classified or something along those lines. The other thing is that AFAIK it's preferable to have even amounts of spam and ham when using sa-learn. I'm sure others have more helpful insights. HTH Dan. -- SLUG - Sydney Linux User's Group - http://slug.org.au/ More Info: http://lists.slug.org.au/listinfo/slug
