I would try mixture model incorporating distributions for categorical data: negative binomial, poisson, etc. This can be considered as latent class analysis - each cluster or latent class would have different distribution.作者: 偏最小二乘 时间: 2011-7-2 22:43