CRP clustering
Summary
The predictive rule for Chinese Restaurant Process (CRP) can be used to define an "infinite-capacity" prior distribution on the clusters in a clustering model. The most common clustering model that uses the CRP is an unbounded analogue to a Gaussian mixture model, where the "table assignments" from the CRP determine the mixture component assignments for each data point.
Context
This concept has the prerequisites:
- mixture of Gaussians models (Mixture of Gaussians is the canonical example of CRP clustering.)
- Bayesian parameter estimation (CRP clustering is a kind of Bayesian model.)
- Chinese restaurant process
- collapsed Gibbs sampling (Collapsed Gibbs sampling is a simple way of performing inference in a CRP clustering model.)
Core resources (we're sorry, we haven't finished tracking down resources for this concept yet)
Supplemental resources (the following are optional, but you may find them useful)
-Free-
→ Bayesian nonparametric lecture notes (COS 597C)
See also
- The Indian buffet process linear Gaussian model is another Bayesian nonparametric model, but which uses latent binary representations in place of clusters.