On Wed, Feb 27, 2002 at 03:30:32AM -0700, Michael Moncur wrote: > I might be wrong, but I think there's something seriously amiss with the new > GA-evolved scores - they don't seem to have an upper boundary (many are 9-10 or > so) or a lower (some are negative). Some examples that can't be right: > > score 25FREEMEGS_URL -4.606 > score BE_AMAZED -4.581 > score CASHCASHCASH -3.700 > score CYBER_FIRE_POWER -4.020 > score DEAR_SOMEBODY -4.412 > score EXCUSE_5 13.447 > score FREE_CONSULTATION 15.263 > score GAPPY_TEXT -3.667 > score IN_REP_TO -13.472 > score MONSTERHUT -8.280 > score ONCE_IN_LIFETIME -4.604 > score PORN_8 -5.452 > score TRACKER_ID -4.899 > > Aside from the boundary issue, is there perhaps something odd about the corpus? > It doesn't include messages from this list, does it? I would think there would > be no trouble calculating a positive score for something like "Monsterhut"... >
Ummm... I'd be heavily inclined to set these spam scores to 0.01. It's not that I don't trust the GA, it's just that if these are the outputs, they aren't needed in the first place. -- Duncan Findlay _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
