On Sun, Mar 16, 2003 at 11:30:38AM -0500, Jay Levitt wrote: > I *think* it has to be 15 points *without* the bayes score. Otherwise, a > high bayes score would simply be self-reinforcing, which isn't learning it > all. > > Hopefully Theo can provide a more authoritative answer.
Heh. Mostly right. What happens is that if you're running in scoresets 2 or 3, when the autolearn question comes up, SA reverts to "$current_scoreset - 2" and rescores the message ignoring "learn" and "userconf" rules. That score is used to determine autolearning. If you run spamassassin -D, you will see something like: debug: auto-learn? safety=4, ham=-2, spam=15, body-hits=4.9, head-hits=1.5 debug: auto-learn: currently using scoreset 3. recomputing score based on scoreset 1. debug: Score set 1 chosen. debug: auto-learn: original score: 6, recomputed score: 5.401 debug: Score set 3 chosen. debug: auto-learn? no: inside auto-learn thresholds or safety zone around required_hits -- Randomly Generated Tagline: "Never doubt that a small group of thoughtful citizens can change the world. Indeed, it is the only thing that ever has." - Margaret Mead
pgp00000.pgp
Description: PGP signature
