Using Empirical Bayes to approximate posteriors for large "black box" estimators
The Unofficial Google Data Science Blog
NOVEMBER 4, 2015
These estimates can be useful to make risk-adjusted decisions and explore-exploit trade-offs, or to find situations where the underlying regression method is particularly good or bad. $mathrm{var}(theta | t, y)$ estimates the accuracy of $E(theta | t, y)$ — this tells us how much we know about each item.
Let's personalize your content