On Sun, Mar 30, 2003 at 10:16:00AM -0800, Bart Schaefer wrote: > Theo remarked in another thread that Bayes is most accurate when the DB > nspam is approximately equal to nham.
That's not what I said. I said that you can't be necessarily surprised
if spam is classified as ham if your training set is heavily weighted
towards ham.
It's really based on nspam/ham per token, but when you heavily weight
in one direction, it's likely the same weighting will occur per token.
--
Randomly Generated Tagline:
First Law of System Requirements:
"Anything is possible if you don't know what you're talking about..."
pgp00000.pgp
Description: PGP signature
