> By learning HAM, am I increasing the probability of HAM tokens?
Perhaps the way to think about it is that by learning ham and spam, you are
showing bayes what ham and spam look like for you, and which is which.
That may translate to probabilities internally, but that is really a level
of detail that is somewhat uninteresting to most people. Think of training
more in the lines of having a guy in the mailroom and telling him "this pile
is junk, this other pile is good stuff. As stuff comes in, decide by
looking at this which pile the new stuff belongs in."
If you want him to do his job of sorting right, you have to give him some
good examples to start with.
Loren
