On Fri, Oct 13, 2006 at 10:09:07PM +1000, Mark Constable wrote:
BTW can anyone suggest a more efficient spam filter other than
I used dpsam in production for a while. It certainly is more
efficient, but it stupidly uses a database for a task that the
file system is much better suited for. If you want to use the
signature feature to train (a requirement unless all users just
use webmail), then plan to put aside _a lot_ of disk space to
store email history.
I have recently been testing out CRM114 Discriminator, and like
it a lot. It's written in C and uses file system, a design that
seems right to me. It supports merged statististics (global and
user-specific), which is what I am working on right now. One
thing I like is the threshold, where it categorizes messages as
"trainme" when it's not sure. I have had very very few false
categorizations--instead, stuff goes into this trainme category.
Development is active and the author is sharp and helpful. The
downsides: complex setup, newer project, and daemon mode with
merged statistics files is virgin territory. I'm also working on
a SquirrelMail plugin to make training easy.