Training spamassissin

[yoja]

I'm having trouble training spamassassin serverwide. (I have one domain of my own which I would like to train and let the all the other domains on the server benefit from my training).

My situations: I have bought the psa-sa plugin and enabled global settings in the control panel. ( Server -> Spamassassin -> Server wide settings tagged
Personal settings not tagged)

I have an spam imap folder in which I move with outlook all my spam. I have an imap ham folder in which I move all my ham. About now there are some 2000 spam and 1000 ham mailes.

I train the spamfilter with a cronjob every night like this:

/usr/bin/sa-learn --spam /var/qmail/mailnames/MYDOMAIN/spam/Maildir/.SPAM/cur

/usr/bin/sa-learn --ham /var/qmail/mailnames/MYDOMAIN/ham/Maildir/.ham/cur

An output of the learing gives something like:
Learned from 46 message(s) (2073 message(s) examined).

But still loads of spam come throug, an email of which I've seen several ones still gets a poorly score of 1.2:

X-Spam-Checker-Version: SpamAssassin 3.0.3 (2005-04-27) on MYSERVER
X-Spam-Level: *
X-Spam-Status: No, score=1.2 required=7.0 tests=FORGED_RCVD_HELO,HTML_MESSAGE,
MIME_HTML_ONLY,RCVD_BY_IP autolearn=no version=3.0.3

It doesn't seem to take in the baysian filtering by autlearning into its filtering.

My questions:
- How do I check if the manual training is used?
- Is my assumption correct that the manual training rules are not taken into consideration?
- How do I enable filtering with my manual training rules enabled.

Hope anyone can help...... The spam is killing me!

 

[fairvegas]

Using sbl-xbl.spamhaus.org to block.

First things, usually being first... have you entered an sbl-xbl into your Server/Mail/MAPS Zone?

Using sbl-xbl.spamhaus.org in the zone is easy and will go along way in at least getting you to a reasonable starting point regarding the blocking of spam.

However, from what I know the training takes place on it's own and once it can begin filtering based on what it's learned, you'll know it. I have an external Barracuda Networks anti-spam/virus server that's based on spamassassin and I've had it for almost 3 years. The most powerful feature is the Bayesian filter, but to activate it, way back when, meant I had to tell it 200 messages that were spam and 200 messages that were good. With all the technologies running on the Barracuda, the Bayesian is by far the most powerful. For instance, I have just over 1,000 mailboxes on my Plesk server over 109 domains and my spam kill rate is 99.8% with no false positives.

 

[benmx]

Hey guys,

I've been battling this spam thing with Plesk for almost a year and I still get upwards of 100+ spam msgs per day. I figure I must be doing something wrong. I just added that url into the MAPS field as suggested... but using Spamassassin on my individual email account hasn't worked a damn. Does using server wide settings make the training moot? I've trained way more than 200 on the spam side and ham side and it seems to be completely ineffective. I know I've got to be missing something... any suggestions?

Ben

 

[pjaric]

Take a look at the email headers and see if atuto ham or auto whitelisting is in effect. You might see negative scores because of this.

I saw this just recently with a client, so I had to clear what they learned and disabled auto hamming and auto white list. Things seem to be going the right way now.