You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Jared Priddy <Ja...@mail.cone.com> on 2004/07/08 23:06:27 UTC

bayes_auto_learn problem

I have the following entries in /etc/mail/spamassassin/local.cf:

use_bayes 1
 
bayes_auto_learn 1
bayes_auto_learn_threshold_nonspam -5.5
bayes_auto_learn_threshold_spam 12.0
bayes_path /var/amavis/.spamassassin/bayes

Since I am only using this as a mail gateway, I have no individual
user_prefs files, this file controls all.  I know bayes is used when
scanning emails since it shows up in the headers, and I can manually feed
the bayes db with no problem, but auto-learn is not working anymore. When I
run spamassassin -D --lint as root, I get the normal stuff except for this:

debug: cannot use bayes on this message; db not initialised yet
debug: bayes: not scoring message, returning 0.5
debug: bayes: 10024 untie-ing
debug: bayes: 10024 untie-ing db_toks
debug: bayes: 10024 untie-ing db_seen

and it does not use bayes. What might be causing this?



Re: bayes_auto_learn problem

Posted by Matt Kettler <mk...@evi-inc.com>.
At 05:06 PM 7/8/2004, Jared Priddy wrote:
>debug: cannot use bayes on this message; db not initialised yet
>debug: bayes: not scoring message, returning 0.5
>debug: bayes: 10024 untie-ing
>debug: bayes: 10024 untie-ing db_toks
>debug: bayes: 10024 untie-ing db_seen
>
>and it does not use bayes. What might be causing this?

This message indicates that bayes could not be used in SCORING the message 
because the bayes database is entirely empty.

This is quite different from anything to do with wether or not SA would 
autolearn the message (autolearning is rather rare). Theoreticaly, even 
with the above message SA could decide to autolearn the message and this 
initalize the database, but so far it hasn't found an email worthy of such 
treatment.

That said, you should never rely entirely on autolearning as your only 
source of training. Badly poisoned bayes DBs are nearly always a result.

Prime your bayes DB with some hand training using sa-learn first. This will 
help prevent bad autolearning, and gives your bayes DB a considerable 
headstart.