Index: The Book of Statistical ProofsModel Selection ▷ Bayesian model selection ▷ Model evidence ▷ Expression using prior and posterior

Theorem: Let $p(y \vert \theta,m)$ be a likelihood function of a generative model $m$ for making inferences on model parameters $\theta$ given measured data $y$. Moreover, let $p(\theta \vert m)$ be a prior distribution on model parameters $\theta$. Then, the log model evidence (LME), also called marginal log-likelihood,

$\label{eq:LME-term} \mathrm{LME}(m) = \log p(y|m) \; ,$

can be expressed in terms of prior and posterior as

$\label{eq:LME-bayes} \mathrm{LME}(m) = \log p(y|\theta,m) + \log p(\theta|m) - \log p(\theta|y,m) \; .$

Proof: For a full probability model, Bayes’ theorem makes a statement about the posterior distribution:

$\label{eq:BT} p(\theta|y,m) = \frac{p(y|\theta,m) \, p(\theta|m)}{p(y|m)} \; .$

Rearranging for $p(y \vert m)$ and logarithmizing, we have:

$\label{eq:LME-bayes-qed} \begin{split} \mathrm{LME}(m) = \log p(y|m) & = \log \frac{p(y|\theta,m) \, p(\theta|m)}{p(\theta|y,m)} \\ &= \log p(y|\theta,m) + \log p(\theta|m) - \log p(\theta|y,m) \; . \end{split}$
Sources:

Metadata: ID: P314 | shortcut: lme-pnp | author: JoramSoch | date: 2022-03-11, 16:25.