bridge: Bridge sampling

In tkmckenzie/ikde: Iterative Kernel Density Estimation

Description

Evaluates marginal likelihood of Stan model at the posterior mean using bridge sampling

Usage

bridge(ikde.model, burn.iter = 1000, sample.iter = 1000, tol = 1e-08,
  control = NULL, refresh = NULL, display.output = FALSE,
  show.trace = FALSE)

Arguments

ikde.model	An object of class ikde.model, does not necessarily have to be built
burn.iter	Number of warmup iterations
sample.iter	Number of sampling iterations
control	Control parameters used in the Markov chain. See ?rstan::stan for details.
refresh	How frequently should progress be reported, in numbers of iterations
display.output	Boolean indicating whether output from rstan::stan should be printed
show.trace	Boolean indicating whether to show trace plots

Details

Uses evaluate.likelihood, evaluate.priors, and a custom bridge sampling routine to estimate marginal likelihood.

Value

A real number indicating value of the log-marginal-likelihood at the posterior mean

Examples

data(lm.generated)

X <- lm.generated$X
y <- lm.generated$y

data <- list(N = list(type = "int<lower=1>", dim = 1, value = nrow(X)),
             k = list(type = "int<lower=1>", dim = 1, value = ncol(X)),
             X = list(type = "matrix", dim = "[N, k]", value = X),
             y = list(type = "vector", dim = "[N]", value = y))
parameters <- list(beta = list(type = "vector", dim = "[k]"),
                   sigma_sq = list(type = "real<lower=0>", dim = 1))
model <- list(priors = c("beta ~ normal(0, 10);",
                         "sigma_sq ~ inv_gamma(1, 1);"),
              likelihood = c("y ~ normal(X * beta, sqrt(sigma_sq));"))

ikde.model <- define.model(data, parameters, model)

# Only an estimation, may not exactly match presented result
bridge(ikde.model)
# [1] -389.0141

tkmckenzie/ikde documentation built on May 13, 2019, 9:53 p.m.