Anybody have any advice on what to set these thresholds at?
Currently I run SA's Spam Threshold at 5.5, the default Bayes numbers are .1 for HAM, and 12 for SPAM.. I put mine to 1.0 for HAM and 6.5 for SPAM.. is that reasonable?
That depends a bit on what your typical spam and nonspam get for scores.
Personally, I'm strongly opposed to the ham threshold being non-negative. I get way too many super-low-score spams for a positive or zero threshold to not cause mis-learning of spam as HAM.
In order to get some good HAM autolearning, I've supplemented SA with a few small negative-score comp rules for terms very specific to the business my company is in. Enough to get a few negative scoring emails, but not enough to be abused by spammers. (most of these custom rules are in the -0.1 to -0.05 range and my learn threshold is -0.1) Even if the spammer hit all of my comp rules in one email, he'd only get a -1.0 or so benefit from it.
-------------------------------------------------------
This SF.net email sponsored by: Enterprise Linux Forum Conference & Expo
The Event For Linux Datacenter Solutions & Strategies in The Enterprise Linux in the Boardroom; in the Front Office; & in the Server Room http://www.enterpriselinuxforum.com
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk