Advice on Using sa-learn

[Matt Kettler]

At 10:04 AM 10/15/2003, Errol Neal wrote:
>Also, these messages,
>since they have been scanned already by mailscanner have additional headers
>added to them. Should
>I configure mailscanner to ignore those message headers when it's using the
>bayes database?

Yes, although the standard spam.assassin.prefs.conf already has this built
in. However, you will need to add them to your user_prefs for when you run
sa-learn. Personally, I just copy spam.assassin.prefs.conf to
/root/.spamassassin/user_prefs, this way root's command-line has the same
behavior as MailScanner.


As for your questions about getting mail to learn from, I have no good
answers for you. The answer to how to use redirect/forward/whatever is
going to be specific to each and every mail client out there.

I see that you're using Eudora, like I do.. and I'll let you in on
something.. Proper learning from mail that has been downloaded to Eudora is
definitely IMPOSSIBLE, no matter what you do.

Eudora drops entire mime sections from the message as it stores them, and
it is impossible to rebuild the correct original message once Eudora
downloads it. For example, any messages with both text/plain and text/html
sections in them will have the text/plain section completely discarded. It
cannot ever be reproduced. Even if you directly convert the Eudora MBX
files, whole mime sections from the original message are permanently lost.

I've had some mediocre success with converting eudora mbx files to unix
mbox format, but due to the differences in format and content I've had much
better success when training spam with real, unadulterated email. To solve
this I've been splitting off some of my own mail at the server side before
eudora pop downloads. However, this doesn't do anything for other users in
the network.