http://bugzilla.spamassassin.org/show_bug.cgi?id=2910





------- Additional Comments From [EMAIL PROTECTED]  2004-01-12 12:11 -------
yep; IIRC, it was either about the same accuracy as bayes alone, or was slightly
worse.  Search the sa-devel archives for "bayes tweaks" for details.

The tests I did were to add the rule names as tokens, and also to add pairs of
rules as tokens (e.g. a message that hit AWL,BAYES_00,NO_REAL_NAME would result
in AWL+NO_REAL_NAME as a token).

One tweak I did not test, however: rules that did *not* hit.  e.g. that message
(AWL,BAYES_00,NO_REAL_NAME) did not hit HABEAS_SWE, RCVD_IN_NJABL_PROXY,
HTML_40_50, and hundreds of other tests; the suggestion was that the *lack* of a
rule hit was equally a useful ham token.

See "masses/bayes-testing/README" for details on running 10fcv runs with SA 
bayes.



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

Reply via email to