carnold5@nc.rr.com wrote:
> I got a message that has tagged as spam. Received a score of 5.2. This
> mail is a ham mail for me/us. So i ran --forget and received this:
> sa-learn --forget --mbox /var/opt/hula/netmail/users/forget
> Forgot tokens from 0 message(s) (1 message(s) examined)
> There was only 1 message/email in this folder. I expected to see Forgot
> tokens from "1" message(s) (1 message(s) examined) but this was not the
> case. What did i do wrong?
> SA 3.2.1 with sles9 and spamd running without any options
>

--forget only works if that specific message has been learned as spam by
the bayes subsystem. And, just because a message is tagged as spam, it
does not mean that the bayes autolearner caused it to be trained.

Really, what you would want to do is --ham, not --forget.

--ham will explicitly add information to the database that the message
is not spam.


--forget will only negate any information resulting from learning that
message, but doesn't change the database in any other way. Generally I
would avoid using forget, it's really a special-case tool only. If a
message was marked incorrectly, feed it to --spam or --ham as needed.

(and no, running mis-learned through --forget first doesn't change
anything. If a message was learned as spam, and you feed it to sa-learn
--ham, SA is smart enough to do a "forget" and "learn as ham" in one pass.)