[Mimedefang] Re: Bayes lockfile problem

Andrzej Marecki amr at astro.uni.torun.pl
Mon Dec 8 05:20:34 EST 2003

Per.Bjorklund wrote:

> The bayes database is quite large on both machines (44MB and 46MB).


> So at the moment I'm forced to disable bayes.. ;(

First off, _what_ option did you disable? To get rid of the problem 
you - and *not only* you - have, it's enough to set:

use_bayes 1
bayes_auto_learn 0

in /etc/mail/spamassassin/sa-mimedefang.cf

And secondly, IMVHO one doesn't need such a large database. I stopped
auto_learn-ing - see the parameters above - when my bayes_toks file was
~5MB large. Still > 90% spams get BAYES_99 and another ~5% get BAYES_90.

Therefore I dare think it's _advisable_ to stop auto_learn-ing at some
level when the Bayesian filter "knows" quite well what spams are all

BTW, as we all prefectly know, BAYES_99 costs 5.4 points. If the spam
level ("required_hits" parameter) is 5, then BAYES_99 alone is enough to
kill such a piece of spam. So it's reeeeeally good to have the Bayesian
filter working! (I'm sure you know it very well and are quite convinced
about it, Per ;-) But not necessarily auto_learn-ing endlessly about
bigger something, cheap/free something etc.

Andrzej Marecki                | 
Torun Centre for Astronomy     |   e-mail: amr at astro.uni.torun.pl
N. Copernicus University       |   WWW:    http://www.astro.uni.torun.pl
ul. Gagarina 11                |   tel: +48 56 6113032
PL-87-100 Torun, POLAND        |   fax: +48 56 6113009
"If Bill Gates actually had to admin his own stuff, he'd shoot himself."

More information about the MIMEDefang mailing list