Brett Dikeman wrote: > I tried setting bayes_auto_learn_threshold_nonspam to a positive > value- almost all legitimate email we get on the particular system is > marked somewhere between 0 and 2- rarely any lower. Ever(save for > whitelisting).
You've found the right setting, and I'm not aware of any particular problems setting to a positive value. (The default value is 0.1.) Staring at the man page reveals no "official" limits in this respect. I vaguely recall that there are a few rules or types of rule that are ignored when calculating a score to apply to decide whether to autolearn, but I don't recall the details. Among other things, however, SA will use one of the two NON-Bayes score sets- which in some cases can result in a *very* different score. > That setting appeared to have no effect. I just sent a test, it got a > score of 0, and was tagged autolearn=no. Even after I removed the > whitelists. I need to change this. Very strange. Even with the default thresholds, I was still seeing messages autolearned as ham on the system I'm maintaining- but I was also seeing spam autolearned as ham. I *dropped* the autolearn threshold for spam down to -0.5 (-2 for a brief time) which seemed to not autolearn spam. In my case, autolearning spam is Really Bad, because I have no way to easily counterbalance such messages. > Also- maybe it's just me, but it seems rather silly to not allow the > user to auto-learn messages that have been whitelisted, either > sitewide or user-specific. Could someone a)explain the reasoning > here I think part of the reasoning is that whitelisting is a last-resort option (before processing outside of SA, of course <g>) to get a message to NOT be tagged. Someone sending legit-but-very-spammy mail whose messages are autolearned will reduce the effectiveness of the Bayes score (by how much I'm not sure) because you're effectively declaring spammy tokens to be less spammy. It would definitely be nice to be able to override this limitation though. -kgd -- "Sendmail administration is not black magic. There are legitimate technical reasons why it requires the sacrificing of a live chicken." - Unknown ------------------------------------------------------- The SF.Net email is sponsored by EclipseCon 2004 Premiere Conference on Open Tools Development and Integration See the breadth of Eclipse activity. February 3-5 in Anaheim, CA. http://www.eclipsecon.org/osdn _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
