Marc Perkel wrote: > As opposed to preprocessing before using SA to reduce the load. (ie. > using blacklist and whitelist before SA) >
One thing I noticed when experimenting with pre-filters: bayes no longer knows about certain kinds of spam. If, for some reason, the prefilter does not catch (i.e. you are one of the first to get a new spam run) then SA might pass it with neutral bayes. So it might be an idea to feed (a certain percentage of) pre-filtered spam to a low priority SA learn job Wolfgang Hamann