SA Bayes Questoin

Nathan Johanson nathan at TCPNETWORKS.NET
Thu Apr 24 21:03:57 IST 2003

OT, but people in this list are so good and responsive (cheap
compliments aside).

I set up a test box with SpamAssassin 2.53 and the latest MailScanner
build. (I wanted to make sure I get all of the potential wrinkles worked
out before going live).

I fired some spam messages through the test system. They do get scored
as spam by Spam Assassin, but the bayes databases have not been created
yet (bayes_msgcount, bayes_seen, and bayes_toks). At this point, I'm
just sticking with the default location of /root/.spamassassin/. Am I
falsely assuming that these databases should be created immediately with
the auto-learning functionality, after only processing a few spam
messages? Or does it take a much larger mass of email before these
databases are created and subsequently populated with information? I
assumed that the first processed spam message would be enough to at
least trigger the creation of the db files.

If I manually run the sa-learn tool against a mailbox...

sa-learn --spam --mbox /mbox_location

it creates the database files and populates them appropriately. A
subsequent run of check_bayes_db does indeed return the expected

I'm just curious how long it takes the auto-learning bayes functionality
to kick in.

Nathan Johanson
Email: nathan at

More information about the MailScanner mailing list