SA-learn no effect

[repasp]

Hello,
I have a pmg spamfilter. The spams are collected on the mailing server and downloaded with a script.
These spam marked emails are given to sa-learn (for example dozens of email from the same sending domain) but it seems it has zero effect on the filtering. I can see the spam filterint headers in the emails, but obvious spam email-s not marked as spam!
What shall I configure in PMG to use the given spam samples to filter spams?
Bayesian filter => yes
RBL checks => yes
Razor2 checks => yes

Any help welcome!

 

[Stoiko Ivanov]

In general we recommend against activating the bayes filter (especially with autolearning) as it skews the results in the wrong direction more often than it helps! (the default of bayes changed to disabled in 8.0, and we recommend this for quite a while longer).

Check out the general recommendations for PMG and better accuracy:
https://pmg.proxmox.com/wiki/index.php/Getting_started_with_Proxmox_Mail_Gateway
These are usually enough to get quite good results

If you still want to use bayes - follow the documentation from SpamAssassin:
https://cwiki.apache.org/confluence/display/SPAMASSASSIN/BayesFaq

I hope this helps!

 

[repasp]

Thank you for your help!
DNSBL are active, and we receive quite a lot of spam. More and more spam.
With bayes off, how can I teach the spamfilter? Our goal is to block contents that is marked spam by users.
I made a script to download the emails from the spamcollector mailbox on mailing server and "sa-learn $email".
If bayes off, I think "sa-learn" is useless. Am I wrong?

The static sender blacklist is pretty ineffective as the spammers can change addresses and domains in seconds.