http://bugzilla.spamassassin.org/show_bug.cgi?id=3447





------- Additional Comments From [EMAIL PROTECTED]  2004-06-14 19:58 -------
Just a few notes in reply to Justin:

1. these are just interim scores for the pre-release, they are (a) better than
   what the perceptron generated before this patch and (b) better than the
   original scores

2. the score ranging code lets scores leap out of the range in the last
   epoch, this needs to be fixed before final perceptron runs are done

3. I'm more concerned about the FN/FP rates we're getting, I think we need to
   do more work on the perceptron, do some 10cfv, etc.

I want to fix the score ranging, find a better solution for the FN rate setting
around 5.0 threshold, and then work on better score ranging algorithms, but I'll
settle for 2 out of 3 before next set of perceptron running.

Still, I think we're ready for a pre-release (we should tell people to ignore
the low bayes scores and just leave them alone, though).





------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

Reply via email to