SpamAssassin and Bayes

[mrezzonico]

Hi all,

I have configured SpamAssassin with Scalix.
It works correctly.

In local.cf I have uncomment the following 2 lines:
use_bayes 1
bayes_auto_learn 1

Moreover as described in the following wiki
http://www.scalix.com/wiki/index.php?ti ... amAssassin
I use a script to fed sa-learn.

Looking in the header of the incoming emails I alway see text like the following::

tests=HTML_MESSAGE,MIME_HTML_MOSTLY,
RCVD_IN_DNSWL_LOW,TVD_SPACE_RATIO autolearn=no

I do not see the string "BAYES_00" after the word "tests", therefore I am not sure if bayes works correctly.

My question is this:
Does bayes works correctly ?
How can I test if bayes works correctly ?

Thanks for any help. Miche

[PrisonMind]

please poste the result of

Code: Select all

spamassassin --lint -D

btw, why you should autolearn???
did you know how you must correct use this option?
the most make many faults with autolearn!!!
standard spamassassin is good enough! or you have special spam!
then you can create easy new rules for this special spam!!
you should better use greylisting,policyd or policy-weight and standard spamassassin!

[mrezzonico]

Hi,

thanks for your reply.

>> btw, why you should autolearn???
The default value of use_bayes and bayes_auto_learn is 1.
Do you mean that I must set this options to 0 ?

By default "spamd" is started as user "root" and therefore is failing back to nobody.
This was the problem.

The solution is to add the following lines to local.cf:

bayes_file_mode 666
bayes_path /var/tmp/spamassassin/bayes

Now I see in the tests that the BAYES-test is done.

Thanks. Miche

[PrisonMind]

many people train only spams with spamassassin and this is shit
you should train both spam and ham

EFFECTIVE TRAINING
http://spamassassin.apache.org/full/3.0.x/dist/doc/sa-learn.html
@ the last mailserver conference in berlin they told that the standard spamassassin is good enough, you should update only the rules or create your own for special spams.

you should test this tool,too
http://www.policyd-weight.org

and rbl's

[mrezzonico]

Thanks

Regards. Miche