Re: [exim] Reducing load vs seeing all the spam

Hi Ted

On 01/06/05, Ted Cooper <eximx1805@???> wrote:
> I use the fakereject and acl varibles along with a special router that writes
> the emails to a spam Maildir folder.

Makes sense.

> Once there, I review it and select which
> mails will be sent to the sa-learn program (Don't want ones that are already
> picked up as 99% spam possibility)

Don't you? I'd have thought you *do* want those, because although a
subset of tests have identified the mail as spam, if you tell SA to
learn the mail it will also increase the likely spaminess of the rest
of the text, thus increasing the accuracy of the Bayesian algorithm
next time.

> which are then processed when the load on
> the server is low.

Yep, I think I'll need to do that.

Actually I've got a bit more of a problem because I'm intending to
share the Bayesian database in MySQL across several servers, and I'll
need to do the 'learning' centrally. So I need to forward the learning
candidates from the satellite servers to the centre.

Thanks for the input...

Peter
--
Peter Bowyer
Email: peter@???
Tel: +44 1296 768003
VoIP: sip:peter@???