From: "Rakotomandimby (R12y) Mihamina" <mihamina@xxxxxxxxxxxxxx> > Hello, > On a dedicated server, I installed spamassassin. > If I change the configuration into /etc/mail/spamassassin/local.cf, > (ie. use terse report or report safe,...) the modifications are applied. > > But there is an exception: the bayes usage. > I know spamassassin uses Bayes by default. If I dont mention any Bayes > related directives into local.cf, it doesn't use Bayes. Though I force > using Bayes into local.cf, there's no way: it doesn't use Bayes. > > spamassassin is run from postfix as a specific user "foo". > I had a look into the user_prefs files of "foo", it doesn't mention any > Bayes, so spamassassin should use it... > > Any other possibility to use Bayes ? > > I alreday ran sa-learn as "foo" many times, and the bayes stuff has been > created under the homedir of "foo". There are potential problems with your understanding, I fear. You must train SpamAssassin with over 200 each of spam and ham before the rules take effect. The way you train is quite dependant on the format for the spam messages used for training. If they are individual files in a directory you give salearn the directory as the parameter. If you have them all in one file in "mbox" format you give that file name as the parameter and MUST include the "--mbox" option in the salearn command. Otherwise you learn the whole batch as a single message. If you've done this your spam database is likely corrupted. Delete it. SpamAssassin will start it over again. The SpamAssassin WIKI is moderately helpful. I recommend it. It will point you towards solutions for the common problems as well as means to enhance its performance. (Of those means I ever so highly recommend the "SARE" rule sets from the SpamAssassin Rules Emporium.) There is an option for disabling Bayes. It defaults to on. Look for information about SpamAssassin as an Apache product at http://www.spamassassin.apache.org/. {^_^}
