Bayesian multidimensional scale clustering based on Dirichlet process
Qing Xiangyun, Xingyu Wang · 2008
An important reason for doing multidimensional scaling is to cluster the objects which are only given dissimilarlity metrics. A prior number of components could be infinite in a Bayesian mixture model. In this work we apply infinite Gaussian mixture model and present a Bayesian multidimensional scale clustering method based on Dirichlet process. Estimating the parameters of Bayesian multidimensional scaling model is done using Markov chain Monte Carlo. As the method avoids the model selection, it can be used not only for generating low-dimensional coordinates and model-based clustering simultaneously, but also for choosing the number of clusters and performing parameters of components at the same time.