I've noticed with SA 2.60, Bayesian autolearning seems to learn a lot of messages incorrectly. As an example, the four spams I've recieved in the last few hours:
% cat spam|grep ^X-Spam-Stat X-Spam-Status: Yes (score: 25.4/6.5), autolearn=spam, version=2.60, X-Spam-Status: Yes (score: 12.6/6.5), autolearn=ham, version=2.60, X-Spam-Status: Yes (score: 23.9/6.5), autolearn=spam, version=2.60, X-Spam-Status: Yes (score: 31.1/6.5), autolearn=no, version=2.60, Possibly relevant lines from my ~/.spamassassin/user_prefs: required_hits 6.5 bayes_auto_learn 1 bayes_auto_learn_threshold_nonspam 6.5 # default -2.0 (0.1 in 2.60) bayes_auto_learn_threshold_spam 6.6 # default 15.0 (12.0 in 2.60) bayes_ignore_header Status bayes_ignore_header Content-Length bayes_ignore_header Lines clear_headers add_header all Status _YESNO_ (score: _HITS_/_REQD_), autolearn=_AUTOLEARN_, version=_VERSION_, tests=_TESTSSCORES_ I do notice the exception to expected "bayes_auto_learn_threshold_spam" behavior in the Mail::SpamAssassin::Conf man page (at least three points must come from the header, and three from the body), but even that can't explain why the second message was learned as ham. I think it would be useful to have a (non-default) mode where simply any message over the user's "required_hits" is autolearn=spam, and everything else is autolearn=ham. Then the user only needs to occasionally intervene (something as simple as a hot-key in their MUA) to move and relearn anything SA got wrong. Any info on the behavior I'm seeing would be appreciated. -- Jeremy M. Dolan <mailto:[EMAIL PROTECTED]> <http://jmd.us/> PGP: 1024D/3C68A1BA 9470 210C A476 FFBB 6D11 0223 0D1C ABFC 3C68 A1BA ------------------------------------------------------- This SF.net email is sponsored by: The SF.net Donation Program. Do you like what SourceForge.net is doing for the Open Source Community? Make a contribution, and help us add new features and functionality. Click here: http://sourceforge.net/donate/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
