Hi, On Tue, 15 Jul 2003 13:40:37 -0600 (MDT) Mike Mills <[EMAIL PROTECTED]> wrote:
> I am experiencing weird and inconsistent behavior with the auto_learn > feature of spamassassin. I set the following in my local.cf file: > > auto_learn 1 > auto_learn_threshold_spam 4.3 > auto_learn_threshold_nonspam 0.0 That's probably not such a hot idea; the auto-learn thresholds are set to (-2 ham/15 spam) for a reason. Better to manually train SA with sa-learn with a corpus of known ham and spam than try to second-guess the autolearning process. > In my user_prefs I have: > > required_hits 4.3 ... meaning all your false positives are learned as spam. :( > It seems that not all of the spam that gets filtered into my spam folder > is learned by the bayes database. Most recently, it seems that spam with > hits of 35.20, 25.10, 23.20, 14.20, and 11.80 points were learned, while > those with 34.8, 34.30, 11.9, 7.7 and 5.00 points were not. Any ideas on > what is going on here? Yes; some scores are not used when calculating scores for the purposes of autolearning. The two that immediately pop to mind are AWL (auto-whitelisting) and the scores resulting from Bayes (avoids degenerate, self-reinforcing behavior.) There may be other scores (DNSBLs?) that aren't used when calculating a score for the purpose of auto-learning. -- Bob ------------------------------------------------------- This SF.net email is sponsored by: VM Ware With VMware you can run multiple operating systems on a single machine. WITHOUT REBOOTING! Mix Linux / Windows / Novell virtual machines at the same time. Free trial click here: http://www.vmware.com/wl/offer/345/0 _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
