With a score as small as -0.036 the GA is giving more of a statement that this rule isn't much of an indicator of anything at all.
I'd agree with the original poster, almost anything coming out of the GA with a score between 0.05 and -0.05 is probably not worth running. Ultimately it is contributing less than 1% of the score needed to reach spam-tag levels (assuming the default of 5.0). grepping 2 different 50_score.cf's out of CVS, SA has 416-632 rules. If a rule is contributing less than 1%, it would take ~1/6th of the total rulebase of similarly scored rules to create a spam-tag. I think it's pretty obvious that such a low-scoring rule is statistically insignificant, particularly given the comparatively low number of total rules. I believe if I created a rule matching the word "news", or other obviously poor indicator of spam, it too might get a similar GA score, which might be an interesting test to run :) I think it would be great for SA to be able to "narrow in" scores within a threshold range to zero. Personally, I go through and hand-edit some of the near-zero entries to zero, since it's obvious to me that the rule in question is not worth 2-3 clock cycles (even if clock cycles are cheap, the rule in question has near zero value). I'd much rather add more rules which are good, strong indicators of spam/nonspam than have lots of rules which really don't correlate well. At 07:28 AM 7/19/2002 +0200, Jesus Climent wrote: >On Thu, Jul 18, 2002 at 05:57:11PM -0500, Shane Williams wrote: > > > > Also, if a single line of yelling scores -0.036, why not just round > > it off to 0 and not have the test run at all? > >Because a single line of yelling seems to be a sign of a legitimate mail >and thus is worth substracting some points to the tag? > >And since the scores are obtained using a GA, that's why the -0.036. > >J ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
