On Tue, 23 Sep 2003 11:41:46 +0200 Federico Giannici <[EMAIL PROTECTED]> wrote:
> I'm using SA 2.60 RC6. > I have noticed that Bayesian scores are not monotonically increasing. > For example, score for BAYES_90 is LESS then the score for BAYES_70 > (when network is used)! > > Is this correct? Yes. Check the mailing list archives; this question seems to pop up every 4-6 weeks or so. > Is there some reason behind it? Yes. All scores are generated by a genetic algorithm (GA) to maximize effectiveness while minimizing false positives (check the docs; 1 false positive is considered ~100x worse than a false negative.) The GA balances scores within those constraints, sometime producing counter-intuitive results. Realistically, any spam that scores as high as BAYES_90 has probably hit many other rules so a lower score for BAYES_90 than for BAYES_70 won't affect overall effectiveness. If I've mispoke, someone more familiar with the GA please correct me... :) -- Bob ------------------------------------------------------- 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
