Hello Matias, Thursday, February 10, 2005, 10:29:08 AM, you wrote:
MLB> Hi, MLB> I'm relatively new using SA and I have a couple of doubts about the MLB> bayes db and the awl db. MLB> I'm running a 3.0.2 site wide install, and I have saw that for each user MLB> there is an .spamassassin directory, storing Bayesian and awl databases MLB> a part from the user preferences file. MLB> The bayes and awl db are working only for the user who owns them right? MLB> What about the bayes data created with the sa-learn command? MLB> I'm being training spamassassin Bayesian filter since the installation MLB> of SA. It sounds to me like you're training a central Bayes database, while the users are using their individual databases. Individual databases are better (non-spam to the business department may look like spam to the bio-chemistry teachers and v.v.), but a central bayes system would work also, but you need to use one method or the other -- teaching a central database which isn't being used by the users will do not good. You should be able to use the command > sa-learn -u $username ... to direct the teaching for each user to his/her database. MLB> The AWL looks like it's activated by default, that could may MLB> cause some problems with the scoring and mark spam as ham and MLB> vise versa? and also could it set wrong scores into the AWL db? Read up on AWL on the wiki. It doesn't store future scores in a database, just records what past scores have been, so it can average out going forward. Yes, its purpose is to bring scores down or up and can cross the spam threshold. If a spam is seen from spammer A with a score of 3, and the next spam gets a score of 6, AWL will bring the score of that second spam down to 4.5. If the next spam is also at 6, AWL will bring the score of that third spam down to 5.25 (this one will be marked as spam). If the next spam (#4) scores at 4.75, AWL will bring that score up to 5.0. Etc. If you have significant false positive or false negative problems, I'd turn AWL off and delete the AWL databases until you resolve those. Once the rules and Bayes systems are working OK, then turn AWL back on. Bob Menschel
