[Mimedefang] bayesian working?
Stephen Smoogen
smoogen at lanl.gov
Sun Apr 11 18:35:21 EDT 2004
On Sun, 11 Apr 2004, Ashley M. Kirchner wrote:
>Stephen Smoogen wrote:
>
>>sa-learn --spam -C /etc/mail/spamassassin --showdots --mbox bad_file
>>sa-learn --ham -C /etc/mail/spamassassin --showdots --mbox good_file
>>
> Okay, so far so good on all replies. Now, next question, again,
>based on the two above commands:
>
> Most, if not all, of my users keep e-mail in separate folders, where
>one of them is their spam/junk folder (assuming they collect them) and
>others for various types of e-mail. Myself, for example, have over 24
>separate folder every month that I file e-mail under. And every month,
>I archive them, and start fresh again (so I end up with
>MIMEDefang-Jan04, MIMEDefang-Feb04, etc., etc.)
>
> Should I run sa-learn on all of these folders and teach it what's
>good, and also on all the spam/junk folders collected?
>
> I realize the more I feed it, the more accurate it can detect spam,
>but at what point is it just too much?
Ok, here are my lessons learned from my bad experience last month :).
1) Get both good and bad emails.
2) Do not get too old of a folder of SPAM/HAM as it will weight older
dates/patterns too high.
3) Make sure you have more than the 1000 emails of both (or change your
numbers in /etc/mail/spamasassisin/sa-mimedefang.cf
Here is my settings for a small site. Make the
use_bayes 1
auto_learn 1
bayes_path /etc/mail/spamassassin/bayes
bayes_auto_expire 1
bayes_auto_learn_threshold_nonspam 0.5
bayes_auto_learn_threshold_spam 10
bayes_expiry_max_db_size 100000
bayes_file_mode 0644
bayes_ignore_header X-Spam-Status:
bayes_ignore_header X-Spam-Score:
bayes_journal_min_size 10240
bayes_journal_max_size 5120000
bayes_learn_to_journal 1
bayes_min_ham_num 100
bayes_min_spam_num 100
--
Stephen John Smoogen smoogen at lanl.gov
Los Alamos National Lab CCN-5 Sched 5/40 PH: 4-0645
Ta-03 SM-1498 MailStop B255 DP 10S Los Alamos, NM 87545
-- You should consider any operational computer to be a security problem --
More information about the MIMEDefang
mailing list