Bayesian modeling in practice
Bayesian model averaging, as encapsulated by Eq. 46, is a very powerful tool for any data scientist to have in their toolkit. In practice, it can take a bit more experience to fully make use of its potential. We haven’t yet said how one goes about computing the expectation value in Eq. 46. This is the practice of Bayesian modeling.
To make Bayesian modeling averaging a practical tool, there are two main approaches we can take:
- Analytical calculation, whereby we approximate the posterior to the extent that calculation of the expectation in Eq. 46 can be done in closed-form or nearly in closed-form, and so we only need to perform a small number of numerical calculations
- Computationally intensive sampling, whereby we numerically approximate the integration in Eq. 46 by sampling many different model values of
We will now cover those two approaches in more detail.
Analytic approximation of the posterior
We have already introduced...
