On Jan 27, 2009, at 11:31, Kai Schaetzl <mailli...@conactive.com> wrote:
LuKreme wrote on Tue, 27 Jan 2009 09:43:15 -0700:
Ah, right, I know that, but that wasn't the question.
No? I thought it was. Or at least in this area. see below.
What I wanted to know is I could use a combination of formail and
procmail to feed each measdsge to sa-learn one at a time.
Converting shell users from mbox to Maildir is not trivial, as far as
I know.
But you wrote your users are Maildir users.
Sorry. Most my users are virual users using Maildir and secure IMAP
(Courier). These users I have no problem with training Bayes.
My shell users are still using mbox. I don't know how easy it would be
to change them to Maildir.
(most of my users are in mysqld/postfix admin with maildirs,
That's why I replied in that fashion. I was assuming you use
Maildir, but
procmail files that spam to mbox folder files.
I haven't ever seen SA segfault on bigger files on my mbox machines. I
remember training my very early Bayes dbs from large mboxes the size
you
mention without a problem.
Yep, so do I.
Those sizes look you waited quite some time
before starting the training.
Until recently I thought all messages were auto learned by Bayes once
I set bayes_auto_learn, so I was not training the low scoring spam or
ham, only the misfiled messages.
Why not just delete and start over and then train daily?
I may have to.
