Discrete parametric graphical models with a Dirichlet type priors
We introduce two discrete parametric graphical models on a finite decomposable graph. They are generalizations of negative multinomial and multinomial distributions. Their conjugate priors are also discussed, which generalize Dirichlet distribution and inverted Dirichlet distribution. Many classical properties of these distributions are generalized, in particular, these new models have explicit normalizing constants and they are graph Markov.
READ FULL TEXT