[EMAIL PROTECTED] (Bob Proulx) writes: > Continuing that thought, I disagree that RBLs should be used by the > GA in SA at all. That RBL data can be different at different times. > I really believe the GA should be trained on the content of the > message without RBL input. If it is deprived of the RBL information > will the GA have improved results over the content? It should. I > believe the RBL input to SA should be a manual control at a low > value. Any individual user can increase the score if they desire.
You need to GA score the RBL rules to achieve a good FP:FN ratio. Without GA scoring of the RBLs, you will raise your FPs too much because the rest of the GA scores are tuned to achieve a good FP:FN ratio. Optimally, we would have different scores for when SA is running local-only versus local+network, but that improvement is much further off (if ever). For 2.41, I did a preliminary pass at using the GA to score the RBL rules (not real-time, but weighted towards recent results, and real-time is the goal) and the real-time accuracy is *much* better now (less FPs *and* less FNs). Some rules got higher default scores and some rules got lower default scores. But, the scores are generally much better. Dan ------------------------------------------------------- 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
