Mailing List Archive

http://bugzilla.spamassassin.org/show_bug.cgi?id=2419





------- Additional Comments From bcwhite@precidia.com 2004-03-24 07:03 -------
Some recent experience on this...

I've actually shifted the weightings somewhat so that the "0" weighting is more
around the 30% mark. I think the settings as I originally listed them are good
_only_ if all the rest of the rule weightings are scored independent of the
Bayes tests. Bayes is then added as an additional correction.

My experience is that a well-trained system ranks HAM almost exclusively in the
0-10% range while a lot of spam comes in at the 40-80% range. A message ranking
40% with a score of 1.0 may be just enough to tag it as spam when combined with
the other rules.

I still believe the weightings originally listed here are good, but (again) only
if the GA generates its scores ignoring Bayes output (i.e. scoreset #1).




------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

http://bugzilla.spamassassin.org/show_bug.cgi?id=2419





------- Additional Comments From jm@jmason.org 2004-03-24 10:35 -------
'My experience is that a well-trained system ranks HAM almost exclusively in the
0-10% range while a lot of spam comes in at the 40-80% range. A message ranking
40% with a score of 1.0 may be just enough to tag it as spam when combined with
the other rules.'

That's very much depending on how many mails you have trained -- a bias in one
direction typically indicates an unbalanced database with more ham than spam (or
vice versa). Also, we had a bug in that dept, too.



------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.