Abstract by Thomas Jensen
Random Focal Partition Process
Statistical clustering models borrow strength within and between groups that can be defined by researchers’ opinion in the Bayesian framework. In the scenario where researchers may have an vague notion of the correct clustering composition, the focal method allows for tractable posterior analysis based on the researchers’ prior knowledge. An extension to the biparametric Chinese Restaurant Process, the focal method has flexibility and tractable normalizing constants that make posterior simulation rela- tively inexpensive computationally. We find our method to contain many advantages over the existing methods.