Bayesian Learning Information
Bayesian classification is a mathematical process through which
machines really and truly can learn. By passing known spam and known
non-spam through a Bayesian classifier our software gets better and
better at recognizing both types of mail. SpamAssassin is able to
utilize the output from the Bayesian learning tool and add/subtract
scores as the rules dictate. Think of this as a way to 'train'
SpamAssassin on what spam and non-spam mail look like. Since we built
the stand-alone command line tool, SpamAssassin can now be trained very
quickly rather than in real time. Overnight SA can be fed months/years
worth of knowledge if you have big spam folders. Each night at midnight,
MDaemon runs an automatic learning program that reads each of the spam
and non-spam messages you have fed it. This way, MDaemon gets a little
bit better in detecting spam and non-spam every day.
The Bayesian process will not kick in until 200 spam and 200 non-spam messages have been fed to it to ensure accuracy.
Additional Comments
For MDaemon versions 7 through 9.08, bayesian scoring will only need 25 messages of each type to begin taking place.