At 12:43 PM 2/17/2004, jdow wrote:
>From: "Mike McMullen" <mlm@loanprocessing.net>
>
> > Hi All,
> >
> > I just installed SpamAssassin recently. I had a spam
> > mailfolder in mbox format but I fed it to sa-learn with
> > out specifying mbox format.
> >
> > I also did this with a ham folder too.
> >
> > How badly does this muck up the database and what
> > do I need to do to rebuild it correctly.
>
>
>I "presume" you have individual ~/.spamassassin directories.
>You will find at least two and maybe three files with "bayes"
>in their names. Delete them and retrain.
>{^_^}
Wiping the entire bayes database, while effective, is quite a bit extreme...
What will likely happen if you forget --mbox is sa-learn will learn the
entire file as if it were one email, and attribute all the tokens to a
single message. Not tragic, but it is innacurate.
If you still have the mbox files you fed to sa-learn, you can just feed
them back to sa-learn with --forget. As long as you feed it with the same
input type parameters, SA should parse it the same way, and correctly
forget all the tokens that it mis-attributed to a single message.
You can then re-train the mbox files using the --mbox parameter and SA will
correctly learn them as multiple messages.