Hi, Currently I testing the packets that contain built-in features for classification. Actually I looked packages such as: e1071, Klar, Caret, CORElearn. However, from what I noticed when building a naive Bayesian classifier, that they package use of the finite mixture model to estimate P (x | C) and using a normal distribution. In my research I use binary data and I want modeled P (x | C), eg the Poisson distribution. Are the packages in the r-project that allows for replacing kernel to estimate P (x | C) as another distribution (the http://www.statsoft.com/textbook/naive-bayes-classifier/)? Or I must implement such a solution yourself? Best Marcin M. -- View this message in context: http://r.789695.n4.nabble.com/Naive-Bayes-Classifier-tp3652658p3652658.html Sent from the R help mailing list archive at Nabble.com.
On Thu, Jul 07, 2011 at 02:18:17PM -0700, m.marcinmichal wrote:> Hi, > Currently I testing the packets that contain built-in features for > classification. Actually I looked packages such as: e1071, Klar, Caret, > CORElearn. However, from what I noticed when building a naive Bayesian > classifier, that they package use of the finite mixture model to estimate P > (x | C) and using a normal distribution. In my research I use binary data > and I want modeled P (x | C), eg the Poisson distribution.Hi. For binary attributes, the distribution P(x | C) is binomial. This is the way, how binary attributes are treated in function CoreModel(y ~ ., lrn, model="bayes") in package CORElearn. I assume that the binomial distribution is used also in other implementations of naive Bayes for binary data. Petr Savicky.