You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Ken Morley <ke...@jmtg.com> on 2004/04/06 22:49:09 UTC

Bayes auto learn not working?

In \etc\mail\spamassassin\local.cf I have the following settings:

bayes_path /etc/mail/spamassassin/bayes
bayes_auto_learn 1
bayes_auto_learn_threshold_nonspam 0.1
bayes_auto_learn_threshold_spam 10.0  

Even though I received some spam in today with scores of 17 and better, none
has been auto-learned.

When I run sa-learn --dump magic, the values I see for nham and hspam are
identical to the number of messages I have manually learned, so I know none
are being auto-learned.

Thanks for your suggestions...

Ken Morley

Re: Bayes auto learn not working?

Posted by Matt Kettler <mk...@evi-inc.com>.
At 04:49 PM 4/6/2004, Ken Morley wrote:
>Even though I received some spam in today with scores of 17 and better, none
>has been auto-learned.

It takes more than just a score over your autolearn threshold to cause a 
message to be autolearned as spam.

         1) The score, recomputed with bayes disabled must be over the 
threshold. This recomputed score is often considerably lower.
         2) There must be at least 3.0 points from header checks.
         3) There must be at least 3.0 points from body checks.
         4) The learning must not contradict any strong existing bayes 
training. (ie: don't learn as spam a message that would have scored 
BAYES_00 previously)

I'd suggest running a message through spamassassin -D. Perhaps the debug 
output will offer some clues...