> Bayes, arrgg!! More than once I've been given examples of bayes being > the solution I need. I really really really want bayes to > work. But each > time I set it up, the db gets huge, scan times go through the > roof, and > I end up disappointed. The one time it worked for me I ended > up training > by hand a couple hours a week, then an hour a week of my > time, then all > on my own time. The result was I caught 20,000 spam a day instead of > 19,500 spam a day.
Using a SQL db for bayes will probably improve it's performance. I got tired of db and expiry problems here too and finally found a way to make bayes work with MS SQL Server. It's been running better since. I set up two IMAP folders for learning and simply drag false positives/negatives into the appropriate folder when one is reported. I also capture some messages based on certain rules they hit, scan them for FPs and learn them. Bayes is working better for us now. Bret
