Bayesian Prior and Posterior Study Guide for ES205 Yu-Chi Ho Jonathan T. Lee Nov. 24, 2000
2 Outline Conditional Density Bayes Rule Conjugate Distribution Example Other Conjugate Distributions Application
3 Conditional Density The conditional probability density of w happening given x has occurred, assume p x (x) 0: N
4 Bayes Rule Replace the joint probability density function with the bottom equation from page 3: N
5 Conjugate Distribution W: parameter of interest in some system X: the independent and identical observation on the system Since we know the model of the system, the conditional density of X|W could be easily computed, e.g., N
6 Conjugate Distribution (cont.) If the prior distribution of W belong to a family, for any size n and any values of the observations in the sample, the posterior distribution of W must also belong to the same family. This family is called a conjugate family of distributions. N
7 Example An urn of white and red balls with unknown w being the fraction of the balls that are red. Assume we can take n sample, X 1, …, X n, from the urn, with replacement, e.g, n i.i.d. samples. This is a Bernoulli distribution. N
8 Example (cont.) Total number of red ball out of n trials, Y = X 1 + … + X n, has the binomial distribution Assume the prior dist. of w is beta distribution with parameters and N
9 Example (cont.) The posterior distribution of W is which is also a beta distribution.
10 Example (cont.) Updating formula: ’ = + y Posterior (new) parameter = prior (old) parameter + # of red balls ’ = + (n – y) Posterior (new) parameter = prior (old) parameter + # of white balls N
11 Other Conjugate Distributions The observations forms a Poisson distribution with an unknown value of the mean w. The prior distribution of w is a gamma distribution with parameters and . The posterior is also a gamma distribution with parameters and + n. Updating formula: ’ = + y ’ = + n N
12 Other Conjugate Distributions (cont.) The observations forms a negative binomial distribution with a specified r value and an unknown value of the mean w. The prior distribution of w is a beta distribution with parameters and . The posterior is also a beta distribution with parameters + rn and. Updating formula: ’ = + rn ’ = + y N
13 Other Conjugate Distributions (cont.) The observations forms a normal distribution with an unknown value of the mean w and specified precision r. The prior distribution of w is a normal distribution with mean and precision . The posterior is also a normal distribution with mean and precision + nr. Updating formula: N
14 Other Conjugate Distributions (cont.) The observations forms a normal distribution with the specified mean m and unknown precision w. The prior distribution of w is a gamma distribution with parameters and . The posterior is also a gamma distribution with parameters and. Updating formula: ’ = + n/2 ’ = + ½ N
15 Summary of the Conjugate Distributions ObservationsPriorPosterior BernoulliBeta PoissonGamma Negative binominal Beta Normal Gamma N
16 Application Estimate the state of the system based on the observations: Kalman filter. N
17 References: DeGroot, M. H., Optimal Statistical Decisions, McGraw-Hill, Ho, Y.-C., Lecture Notes, Harvard University, Larsen, R. J. and M. L. Marx, An Introduction to Mathematical Statistics and Its Applications, Prentice Hall, 1986.