Matthias Haegele wrote: > Bowie Bailey schrieb: > > Grant Peel wrote: > > > I have not turned on bayes, is it critical (to catching spam)? > > > > Bayes is not critical, but it can be very useful. For best > > results, I suggest you do this: > > ACK. It can kick the spam over the treshold which is maybe not hit by > other rules, a well trained Bayes is essential i think. (And it > produces no false-positives if bayes was: BAYES_100 it was always > right).
Very useful, yes. Critical, no. > > Manually train the Bayes db with hand-sorted ham and spam at least > > until you get to the 200-ham/200-spam limit. After that, keep an > > eye on your incoming mail and retrain any messages that are > > mis-classified. > > > > Manual training works like this: > > > > sa-learn --ham /directory/with/nonspam > > sa-learn --spam /directory/with/spam > > You should run sa-learn with the proper user account e.g.: > > > sudo -u amavis -H sa-learn --spam /path/to/spam-messages/ Thanks, I forgot to mention this piece of (critical) information. Training the wrong database is, unfortunately, a common problem. > > By default Bayes will also auto-learn incoming messages as either > > ham or spam based on certain criteria. Some people suggest > > adjusting the criteria to further prevent mis-training, but I have > > not had any problems with the default settings. However, on some > > of my accounts, I will disable the autolearning and manually sort > > and learn on all of my incoming mail each day. > > Autolearning is not failure proof i think, > especially on less restrictive Mailinglists ... It's not failure proof, but most of my accounts rely on autotraining with the default settings. I have had one instance where I had to recreate a corrupted database, but this was on an account that had no manual training done at all. -- Bowie
