* M. Fioretti <[EMAIL PROTECTED]> [2007-10-05 08:19]: > On Fri, Oct 05, 2007 07:56:44 AM +0200, Christian Kuka ([EMAIL PROTECTED]) > wrote: > > At the moment I'm using bogofilter, razor, pyzor, dcc, spamassassin > > and clamassassin and the following procmail rules: > > This (cascading several filters, lightest to heaviest) is what inmy > other reply I said I'd like to try someday. Do you have any _measure_ > of how better it is wrt a bogofilter-only setup? > > Something like "bogo alone stops 80% of spam, razor 80% of bogo > missed, SA 80% of what is still left etc".
I did that myself for quite some time, but note that while this of course reduces the number of false negatives, it also increases the number of false positives, so you won't necessarily increase the global accuracy with such a setup. Giving useful _measures_ is really hard though as they will heavily depend on the configuration/training of the filters and of course also on the actual mail corpus they operate on. Holger
