evaluate.posterior: Stan model posterior evaluation

In tkmckenzie/ikde: Iterative Kernel Density Estimation

Description

Evaluates posterior of Stan model at specified evaluation point

Usage

evaluate.posterior(ikde.model, eval.point, burn.iter = 1000,
  sample.iter = 1000, 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
eval.point	A list of parameter names and the point to evaluate the posterior
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 list of ikde.model objects created by create.restricted.models to estimate posterior density. Each ikde.model is fit, then conditional posterior density is estimated at the specified point.

Value

A real number indicating value of the log-posterior at the specified evaluation point

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)
ikde.model <- build.model(ikde.model)
stan.fit <- fit.model(ikde.model)
stan.extract <- rstan::extract(stan.fit)

eval.point <- list(beta = apply(stan.extract$beta, 2, mean),
                   sigma_sq = mean(stan.extract$sigma_sq))

# Only an estimation, may not exactly match presented result
evaluate.posterior(ikde.model, eval.point)
# [1] -1.889711

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