When dealing with homogeneous types of data, we can employ mixtures of gaussians (for continous) or some kind of k-proptotyping (ordinal-nominal).
I am investigating around the statistical/machine learning comunity and found libraries in R such as clustMD or MixAll that can deal with heterogeneous probabilistic clustering.
However, the algorithms typically need some kind of Gibbs sampling or MCMC, this makes them incredibly expensive in computing terms.
I really need to go into bigger data with those kind of algorithms.
Does anyone have tips for this type of work, knows any library that is powerful in R or Python, with good multiprocessing or gpu applied?
Also, the last decade has led to plenty of work in correlation clustering and cópula formulas. I really don't want to go through a Naive-Bayes assumption in my model.
Am I asking for the impossible?