On Sat, 12 Sep 2009 17:27:00 +0100 Clunk Werclick <mailbacku...@googlemail.com> wrote:
> On Sat, 2009-09-12 at 08:54 -0700, John Hardin wrote: > > On Sat, 12 Sep 2009, Clunk Werclick wrote: > > > > > On Sat, 2009-09-12 at 16:15 +0300, Jari Fredriksson wrote: > > >> > > >> What's wrong with the bayes? > > > > > > Bayes is going out of fashion. > > > > Since when? And according to whom? Bayes is one of the stronger > > tools available. > I disagree. It can do as much harm as good. My own view and > observation from the past have rendered it pointless in my context. > It adds latency, is easily poisoned and rarely makes much difference > to the score. I do appreciate some people like it, but my own view is > spam has moved on beyond the point of it being useful. Unless you're importing porn and viagra from Nigerian lawyers, I doubt your circumstances make the difference between "one of the stronger tools" and "pointless". So you might entertain the possibility that you're are not doing it right. > > > It's just as easy to make a bad one by bad training than a good > > > one. > > > > Any system can be rendered useless by mismanagement. That's not a > > flaw of the system, or a reason to discard it as pointless. And > > GIGO will never become obsolete. > > > Set up bayes and make the commitment to train it properly and > > you'll get good results. > No thanks, I'll pass on that. In this specific case it still would not > have increased the score to a point where the clock cycles made it > worth it. I doubt that's significant compared with the thousands of regular expressions that SA runs. If Bayes slows down SA it's usually a database problem.
