> I came across a situation that seems non-intuitive; > > Two emails this am were spam, but hit BAYES_00. So they were > (presumably) learned as Ham somewhere along the way.
Not a valid presumption. The tokens may have been learned as ham from other messages, but there is no implication that this particular message was learned as ham. > > So far so good... > > Doing ' sa-learn -forget ./message.txt ' gets me : Forgot > tokens 0 from message(s) (1 message(s) examined) > > What kind of situation can cause this ? I was under the > impression that Bayes_00 meant it was explicitly learned as > spam, so there must be related tokens. So this particular message hadn't been learned at all. How about learning it as spam instead? Bret
