SpamAssassin Bayes filter and MailScanner

[Mike McMullen]

> At 06:00 PM 2/16/2004, Mike McMullen wrote:
>
> >I fed about 1000 spam emails today to sa-learn. Watching the
> >mail logs for MailScanner entries, I never see the Bayes Score
> >show up.
>
> How many ham emails did you feed to sa-learn?
>
> Bayes will NOT activate without 500 spam and 500 ham messages, minimum.
>
> Also, beware that spam.assassin.prefs.conf makes mailscanner's
> bayes_path  different
>
>          bayes_path                 /var/spool/spamassassin/bayes
>
> Thus, when you train, you need to train with an account that has that same
> bayes path in it's user_prefs (ie: I do this with root).
>
> Also make sure that /var/spool/spamassassin exists as a directory.
>
> DO NOT make a subdirectory named "bayes" in there.
>

Matt, this is all good info. I have not fed it any ham mail. Now I see
I'll have to do that.

In my MailScanner.conf, the bayes_path is commented out. It is using
/root/.spamassassin which is fine by me. I use root to do the training.
I'm assuming that all the results are getting stored there and not just
mail to root. Is this true?

So if I feed about 500 - 1000 ham messages to sa-learn, bayes will kick
in?

If I have a large number of ham messages in mail folder created by
Outlook Express, can I copy that to the server and then just feed it
to sa-learn?

I did that with sa-learn for the spam messages. sa-learn kicked back
that it had read one message. Is that an appropriate respones? The
bayes_toks database grew to about 2.75M so I assumed it did the
right thing.

Thanks for all your help on this,

Mike