Martin Radford writes: >I've been running 2.60-pr1 since this morning (thanks to Theo for his >speedy assistance with a Perl 5.005-related bug). I've noticed that >nearly all the spams I've had so far have come up with BAYES_99 and >that the probability is given as 1.000. > >With 2.55 I didn't get BAYES_99 as often (though spam did generally >get BAYES_70 and above), and the scores weren't very much higher than >0.990. So I was surprised to see so many getting 1.000. > >My question is: is this expected behaviour? Is Bayes in 2.60 that >much better than in 2.5x?
It's using a scheme called chi-combining, which is more efficient at combining Bayes scores into a "correct" diagnosis than the traditional "gary-combining" scheme. As a side effect, the results cluster more around 0.0, 0.5, and 1.0 than the old classifier did. --j. ------------------------------------------------------- This SF.Net email sponsored by: Free pre-built ASP.NET sites including Data Reports, E-commerce, Portals, and Forums are available now. Download today and enter to win an XBOX or Visual Studio .NET. http://aspnet.click-url.com/go/psa00100006ave/direct;at.asp_061203_01/01 _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
