When NB fails, it is usually due to over-fitting because the training data is relatively small, not because the prior is ignored.
See Rennie's paper for more discussion. http://people.csail.mit.edu/jrennie/papers/icml03-nb.pdf Can you say more about your data size? 2012/1/4 enyun <[email protected]> > hi all, > > I'm trying to use the mahout bayes method to solve some classifier problem. > But I found the result is very bad. > When I dived into source code, I found the prior class distribution was > not considered into model ( for example 20-news case ). > Was it supposed to do like this or a bug here? > > P( c | d ) = p(c) * p(d|c)/p(d) = p(c) * p(t1|c)*p(t2|c)***p(tn|c) / p(d); > here, the p(c) was ignored in real prediction progress. > > thanks, > enyun >
