Well, maybe not dead... What's our goal for the two implementations of Naive Bayes (and Complementary)? It seems to me like the old one, o.a.m.classifier.bayes, is intended to be deprecated due to the fact that it is tied to a word based representation. However, it seems to still have a few interesting things (perhaps) that o.a.m.classifier.naivebayes doesn't (e.g. pluggable model storage.) I still have an open question on the quality of the latter, too, when it comes to many labels that needs to be worked through.
In the spirit of cleaning up, are we going to axe the old one? -Grant