if i'm manually training (don't ask why), does it make any sense to train on messages that score 1.0? is there a sensible cutoff, beyond which i can say "this thing really knows this is spam, i don't have to tell it."? or, is it better to train on all spam, even if it knows that it's spam. ...atom _______________________________________________ PGP key - http://smasher.suspicious.org/pgp.txt 3EBE 2810 30AE 601D 54B2 4A90 9C28 0BBF 3D7D 41E3 ------------------------------------------------- "If you step back and look at the data, the optimum amount of red meat you eat should be zero." -- Walter Willett, M.D., of Brigham and Women's Hospital, director of a study that found a close correlation between red meat consumption and colon cancer.