Non-Parametric Bayesian Methods
Building a predictive model requires us to make assumptions. For example, we often need to assume some fixed mathematical form for the relationship between our predictive features and the response variable. It is the parameters within that mathematical form that we usually vary and optimize through a training process, not the mathematical form. If those parametric assumptions are incorrect, we get a poorly performing model. Often it would be better to not make those parametric assumptions and to use a non-parametric modeling approach. That is what we do in this chapter. We do so by putting Bayesian priors on the functions and relationships that we model. This makes the methods we use non-parametric Bayesian methods. To learn about them we must introduce some new modeling ideas and concepts. We do that by covering the following topics:
- What are non-parametric Bayesian methods?: This is where we learn about the key concept of not making parametric...
