PDA

View Full Version : Using Bayes in Spamassassin



seaquest
08-28-2004, 09:15 AM
With Bayes is it also required to turn autolearn ON? I've just had spamassassin reloaded, it was giving me all sorts of problems for the past several months. I'm wondering if it was because autolearn was left on and the threshold was either too high or low. Is there a risk to using Bayes with autolearn?

FZ
08-28-2004, 09:23 AM
I think the two go hand in hand. If you disable Bayes, then autolearn is disabled automatically. If you have Bayes enabled but autolearn disabled, then I think it is up to you to manually "teach" SpamAssassin which mail is spam and which is ham.

By the way, you can change the threshold at which mail is learned as spam (or ham) in user_prefs. Just have a look at the SpamAssassin docs.

seaquest
08-28-2004, 09:30 AM
Hi Fayez,
I know I can adjust the threshold. Before spamassassin died and I got it reloaded, it got to the point that I needed the threshold at -4.0 to catch much of anything. It was pretty good at that level but it seemed strangely low since most people on here talk about 4.0 (positive) being typical. I was wondering if running autolearn incorrectly for a long time had somehow skewed my numbers.

FZ
08-28-2004, 09:44 AM
I meant the threshold at which mail is learned. For example, you could set it so that any mail with a score of -5 or lower is learned as ham. On the other hand, anything with a score of 5 or higher could be learned as spam. These two thresholds are independant of the spam threshold (that could be anything you like).

A "general spam threshold" of -4 is very, very low - yes. So, your numbers probably were messed up somewhere. As you said, 4 or 5 is a better value to use. Of course, you can always reset your database if it causes problems (I think all you need to do is remove the bayes_* named files in your <homedirectory>/.spamassassin folder).