On Sat, 06 Mar 2010 11:31:17 +0100 Kai Schaetzl <mailli...@conactive.com> wrote:
> Rw wrote on Sat, 6 Mar 2010 01:04:20 +0000: > > > There's nothing odd about that, it's common that hard to learn spam > > is identified correctly on retesting. > > I'm not sure what you want to say. Do you want to say that a message > tested right after learning may get 99, but next day it will have 50 > again? No. I mean that a similar spam from the same spammer that's not been learned will hit 50. The way that bayes performs on retest isn't a reliable predictor of how it will perform on subsequent new spams. The fact that a single spam went to from 50 to 99 doesn't imply that the rest of the that type of spam is easy to learn.
