Rob Mangiafico wrote: > On Tue, 2 Oct 2007, [iso-8859-2] Micha³ Jêczalik wrote: >> There are many. It allows you to share data between user accounts (IMHO it >> doesn't make much sense to have separate bayes databases for each account, >> at least they are of a 'massive' sort and users are not allowed to feed >> their own spam/ham etc. - because they share mostly the same data and the >> bayes is more up-to-date if one single database autolearns from many >> mailboxes). It allows you to share data among several hosts. It allows >> you to keep data on a remote host if you don't have enough space. Etc. > > Picking up on the point of one Bayes DB in MySQL vs. individual ones for > each user, is it more effective in an ISP/host environment where you have > diverse users to have them all share one Bayes DB with autolearn, or is it > better if they each have their own Bayes data in MySQL (per user)? > > We're slowly converting to mysql for bayes, and have not decided yet which > method would be best for our users and for the servers in general. Thanks. > > Rob > >
Per-user Bayes should be more accurate for each user assuming assuming the user can train false positive/negative, using the spam button to "unsubscribe" doesn't impact other user's accuracy. However there is a significant storage cost of per-user Bayes.