On Sun, Aug 17, 2008 at 10:48 PM, David C. Rankin <drankinatty@suddenlinkmail.com> wrote:
Now you have raised another question - Since I do maintain my bayes_toks file and move it from box-to-box and install-to-install, would it be better to clean it up with unlearn instead of simply whitelisting? I guess it really doesn't matter as long as other characteristics of the offending tokens are not causing problems with other mails by increasing their score? I don't know enough about the type and number of "tokens" spamassassin takes per message. What are your thoughts?
Whitelisting is pretty bullet proof, and essentially bypasses any bayes tests, and is suitable for specific addresses of senders that are not likely to be faked. That said, I've never found a reason to bother using Whitelists. With a properly trained bayes (trained with BOTH spam and ham) its just not a problem. Using the two training mailbox scenario (talked about earlier in this thread), you can train bayes with spam and ham by dragging properly categorized mails into the corresponding box to confirm and enhanse bayes recognition in the future. Example: My airline emails look sort of spammy, but were initially passes as Ham (fairly high scoring ham, but ham never the less). I copied them into the Not Spam folder, which got fed to sa-learn and subsequent to that they scored much lower (more hammy) and I've never had a problem. -- ----------JSA--------- Someone stole my tag line, so now I have this rental. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org