Hi. I'm running an email server that's set up largely according to these instructions (Postfix/Dovecot/MySQL/SpamAssassin). One thing I'm having trouble with though is training the Bayesian spam score thing based on the spam that I get.
I'm running commands like:
sudo sa-learn --spam -u spamd --dir /var/mail/vhosts/domain.com/virtualusername/.Junk/* --dbpath /home/spamd/.spamassassin -D and
sudo sa-learn --ham -u spamd --dir /var/mail/vhosts/domain.com/virtualusername/.INBOX.Accounts/* --dbpath /home/spamd/.spamassassin -D And these seem to run fine. For a while there, it was really working... most spam was getting auto-moved to the spam folder (I set up a Dovecot sieve to do the auto-moving if SpamAssassin flags something). But today I did it, followed by a sudo sa-update just in case and it seems like things are much worse now, with more Spam getting delivered.
I don't see anything too useful in the logs. So I have a few questions:
- Do my commands seem right?
- What permissions shouldthe various spamassassin locations have (e.g.
$HOME/spamd,/etc/spamassassin, etc. - What could cause a sudden drop in Bayesian filter quality? (Besides training on the wrong folder... I'm sure I didn't do that).
If it matters, this is a personal server so I thought I'd be ok putting all spam tokens in the ~/spamd folder.
[link][3 comments]