I've been using spamassassin for a while but have never done much with the bayesian component of it. My setup is on a YellowDogLinux box, with Postfix using MySQL authentication and Courier-IMAP. I use maildrop to handle filtering, and spamassassin is called through maildrop for the users/domains that that allow to use spamassassin. For my own usage, I rely on Horde/IMP pretty much exclusively as my email client. I'm using the latest (v2.63) of spamassassin.
Question is how to best use the bayesian component of spamassassin. When I first installed SA, I just did enough to get it going and then left it. From time to time, I would use the "report as spam" option in Horde/IMP without noticing any difference in spam filtering. I have that command set to run "/usr/bin/spamassassin -r". Reading up on the SA FAQ regarding bayes, I see that this is only half the solution. "learn" is the rest of it. I'm not sure about my implementation though. I have postfix using maildir files. Since I use mysql authentication, I can't use procmail per user. So what's happening to the spam i "report as spam" in Horde/IMP? What's the most transparent/automated way of doing "learn"? By now I am sure I have reported more than 100 spams, but not necessarilty 100 of the same variety of spam. What about ham? I have all my legitimate mail sorted according to topic, so there is no single folder that includes all my ham. Is learn on a user-by-user basis, or will running learn "educate" the whole SA installation? Any answers are greatly appreciated. Rob
