Theo Van Dinter writes: > On Wed, Sep 13, 2006 at 09:08:02AM -0700, benthere-nine wrote: > > No one ever said the Bayes scan would be perfect, but > > that's what we want. > > Patches welcome. ;)
One thing I was thinking of was plugins to supplement Bayes with other forms of machine learning algorithms. For example, logistic regression, Andrej Bratko's PPM-D compression classifier, and others seem to be coming out with consistently good results in recent testing -- there's no need for SpamAssassin's only user-trainable machine learning component to be Robinson-style Bayes, if it's possible to get good results with other systems too ;) --j.
