> From: Charles Gregory [mailto:[EMAIL PROTECTED] > Is there really any reason to keep testing a piece of mail once we > know it is spam?
Indeed not. >From a purely efficiency-oriented point of view, the best approach to take is the following: 1) Run all tests that can potentially produce a "ham score" - that is, a negative spam score. 2) Run remaining tests as necessary, "most useful" first. I define a usefulness index as follows: Usefulness = (Potential Benefit) / (Potential Cost) Potential Benefit is easily defined as the number of spam points the test would award if successful. Potential Cost is more abstractly defined as "well, if the DNS lookup times out, that's 30 seconds of wasted time... but it's asynchronous... so... well, there's a cap on the number of threads... I'd need to extract a 20 MB zip file..." etc. Basically run all the 5.0 $Subject =~ /Viagra/ tests first. Save the 0.1 eval:complex-ai-program.pl tests for later in case you need them.
