In joepowers16/rethinking: Statistical Rethinking book package

Bayesian models learn from data.

dunif(seq(0, 1, by = .1), 0, 1)

Imagine that some binary event has a true probability (p) of occuring. But before we know p's true value (if we can ever know it), there are many possible values of p that we'd consider plausible.

Maybe we have no prior knowledge of p and consider all values equally plausible, in which case our prior for p is a uniform distibution between 0 & 1. As we gain more data, there become relatively fewer or relatively more ways that different values of p could produce the data. Values of p with more ways to produce the data are more plausible, and thus we begin to use more informative priors.

Keep in mind what a Bayesian model is asserting: The most probable reality (e.g., the real value of p) given the data. MAP estimates say, given the data, the peak of my distribution is the most probable parameter describing the process that produced the data.

joepowers16/rethinking documentation built on June 2, 2019, 6:52 p.m.