Use scikit <http://scikit-learn.org/stable/> - we've found that it works better than the classifiers in nltk. For the spam/ham problem, I've heard (circa 2006!!) that naive bayes works as well as any other classifier. For the label classification problem, since there is dependency between words, some other classifier will probably out perform a naive bayesian.
hth, Vijay -- Performance marketing on Twitter - http://www.wisdomtap.com/ _______________________________________________ BangPypers mailing list BangPypers@python.org http://mail.python.org/mailman/listinfo/bangpypers