Initial feed for spam filter

[michabbs]

I have large collection of old spam (and much larger storage of "ham"). How can I use it in order to "teach" newly configured PMG? What is "the correct way" to provide PMG with samples of spam/ham?

 

[Stoiko Ivanov]

PMG's spamdetector is (using) SpamAssassin - the learning component of SpamAssassin is its Bayes implementation:
https://cwiki.apache.org/confluence/display/spamassassin/BayesInSpamAssassin

PMG uses the local database of user root for it's bayes db - so you should run sa-learn on your mailboxes as root (make sure Bayes is enabled before in the GUI->Configuration->Spam Detector->Options)

If possible please let us know how this affects your detection rates
Thanks!

 

[michabbs]

Is it enough to do it on one node only?

 

[Stoiko Ivanov]

Is it enough to do it on one node only?

no the bayes-db is not synchronized in any way - you need to run this on all your nodes (or learn on one node, stop pmg-smtp-filter, and sync /root/.spamassassin to the other nodes.

I hope this helps!