Thanks, that handles the top level. ;)
Yeah, it was quick and simple for just the one scenario you had in your
e-mail.
Me, I redirect mail using a combo of procmail and Postfix header checks to
2 users on the border servers (hamfilter and spamfilter), then I do 2
nightly script runs to sa-learn ham and spam. I feed somewhere around
6,000 spam e-mails alone nightly to sa-learn. Maybe that's a bit much,
but I get awesome results, and my FP rates are next to nil. Mind you, I'm
doing this site-wide on border servers, we pass 30k e-mails daily through
those particular systems.
I figure I'll need to do something like:
find mail/Lists -type f -exec sa-learn --ham --mbox {} \;
(I'd need the same for mail/Friends and a few other top-level hierarchies,
excluding my mail/Spam one. Within that tree, I need to put SpamAssassin and
Uncaught under --spam and FalsePositives under --ham.)
Well, another way you could do it is just keep a text list of your spam
and ham folders -
ham.txt:
mail/foo/hambox1
mail/bar/hambox1
spam.txt:
mail/foo/spambox1
mail/bar/spambox1
Then, the original for loop would work:
#!/bin/sh
for x in `cat ham.txt`
do
sa-learn --ham --progress --mbox $x >> outfile
done
cat outfile | mail [EMAIL PROTECTED]
#!/bin/sh
for y in `cat spam.txt`
do
sa-learn --spam --progress --mbox $y >> outfile
done
cat outfile | mail [EMAIL PROTECTED]
But I want to exclude my .imap folders created by the dovecot IMAP server to
hold state data. I might also need to wrap sa-learn in a script to lock the
mailboxes against modification by dovecot and procmail (my LDA).
To build the original text files, you could use find, or edit by hand.
This way you could build a list of your mailboxes, and you can
include/exclude whatever you want.
If you have those boxes as active, then yes. But then again, if you learn
a mailbox that you used to learn before, then it's a waste of cycles for
the mails sa has already seen.
And what would be the equivalent for mass-checks?
Don't use those, sorry...
-Gary
