On Sat, 2006-02-18 at 14:39 +1100, TongMaster wrote: > On Sat, 2006-02-18 at 12:59 +1100, Erik de Castro Lopo wrote: > > > Thats really nice, but it seems a little strange to have everyone > > subscribed to list all devoting cpu cycles to filtering if a single > > filter on the list input would serve the same function.
Seconded. > The filters on a centralised server will always be behind current SPAM > trends, as far as I can see because it's more difficult to train than > your local SPAM solutions. Yes, we should have a centralised SPAM filter > (and probably do) but without someone training it it's going to lag and > more SPAM will get through. Currently spamassassin via a mailman plugin, and a handful of header checks in mailman. A little while before maddog was decommissioned, it hosed SAs Bayesian databases. That was causing problems in the interval before moving to rusty, and meant that we had to start over with a clean slate. Earlier in the week I threw a couple month's worth of my own personal spam through sa-learn. But I haven't had time to sort out a good corpus of ham to train the filter on until, well, a couple of hours ago. Chris also reminded me about James Gray's awesome list of custom spamassassin rules ( http://files.grayonline.id.au/ ). We've been using these for a while, but I just plain forgot to add them on the new box. Thanks again for maintaining these, James. Ongoing training is much less of a hassle; I use a shell script that will ssh to the server doing filtering and run sa-learn, and have been piping wayward spam through that with evolution filters. If anybody would like to contribute something to SLUG, and knows their way around python, there's a couple of things that would be useful: - We're currently using James Henstridge's spamassassin plugin for mailman ( http://www.jamesh.id.au/articles/mailman-spamassassin/ ). Everybody has a favourite spam filter. And being able to plug one or two more in to mailman would be neat. - A patch for mailman to let list admins train the database directly from the moderation page would rock. On the few occasions I see false positives held for moderation I've been forwarding them to myself and training the database by hand. It's a hassle. -- Pete -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
