Hello Burton, Friday, September 16, 2005, 6:22:54 PM, you wrote:
BW> I'm a fairly new user of SpamAssassin, running Debian's 3.0.4-2. BW> If I get a piece of spam that, for some reason or another, gets a BAYES_00 BW> score and nothing else, it will score negative points and is auto-learned BW> as ham. BW> To keep my Bayes db is line, do I have to sa-learn --forget this message BW> and then sa-learn --spam this message, or will just learning it as spam BW> unlearn it as ham? As you've already read, --spam is good enough to retrain Bayes. Assuming you're also manually training ham, I suggest lowering your auto-learn ham threshold down to -0.1 -- don't learn anything which scores zero or positive as ham -- require that it hit some negative scoring rule. If you don't have any negative scoring rules, you'll need to manually learn your ham to keep your Bayes up to date with the ham coming into your system. Bob Menschel
