Dirichlet processes
GPs are not the only type of process used in non-parametric Bayesian methods, although they are possibly the most used. In this section, we will introduce another type of stochastic process, the DP. As with GPs, we will use DPs as priors for functions that we want to make inferences about.
Since the last section was a lengthy one with a lengthy code example, we will keep this section short and only give a high-level view of DPs.
How do DPs differ from GPs?
As you might have guessed, we use a DP as a prior on a function. However, GPs already did that for us, so how do DPs differ from GPs? When we used GPs in GPR, the GP provided a prior for a generic function, 
. There were no restrictions on the type of function, , could be. Sometimes, we will want to model a particular type of function. For example, we might need to build a model of a probability distribution. In this case, we use a DP to construct our prior.
Let’s make that more explicit. I...
