SBC: SBC class

In jasonmtroos/sbcrs: R package for implementing simulation based calibration using rank statistics

Description

Generator class for creating new instances of the SBC R6 class.

Usage

SBC

Format

An R6 object of type SBC

Details

See vignette('intro-to-sbc', package = 'sbcrs') for an accessible introduction to simulation-based calibration.

Methods

$new(data, params, modeled_data, sampling)

Create a new SBC object, passing in functions to generate data and parameters, and draw samples.

data = function(seed) {}

A function with signature function(seed) that returns a named list.

params = function(seed, data) {}

A function with signature function(seed, data) that returns a named list.

modeled_data = function(seed, data, params) {}

A function with signature function(seed, data, params) that returns a named list.

sampling = function(seed, data, params, modeled_data, iters) {}

A function with signature function(seed, data, params, modeled_data, iters) that returns a stanfit object run for iters sampling iterations.

$calibrate(N, L, keep_stan_fit = TRUE)

Run the calibration procedure.

N

The number of times to simulate parameters and recover via MCMC.

L

The number of MCMC samples to retain when calculating rank statistics.

keep_stan_fit = TRUE

If TRUE (the default), then the stan_fit objects returned by the sampling function are retained.

$summary(var = NULL)

Summarize results of a previous calibration. Optionally specify a parameter var.

$plot(var = NULL)

Plot a histogram of ranks from a previous calibration. Optionally specify a parameter var.

Fields

$calibrations

A list of N calibrations created by calling $calibrate. Each item is list with the following named elements:

data

Output from call to the data function.

params

Output from call to the params function.

modeled_data

Output from call to the modeled_data function.

samples

A stanfit object returned from call to the sampling function if keep_stan_fit = TRUE, otherwise NULL.

ranks

A named list matching items in params. Values express the number of samples out of a maximum L with sampled$var < param$var, where sampled$var indicates a vector of L samples of parameter var.

n_eff

The smallest effective sample size for any parameter in any of the stan_fit objects.

iters

The number of MCMC iterations (before thinning) in the stan_fit object.

seed

The value of seed passed to the data, parameter, and sampling functions.

Examples

## Not run: 
sbc <- SBC$new(
  data = function(seed) {
    list(n = 10)
  },
  params = function(seed, data) {
    list(mu = rnorm(1))
  },
  modeled_data = function(seed, data, params) {
    list(y = rnorm(data$n, mu, 1))
  },
  sampling = function(seed, data, params, modeled_data) {
    stan_object <- NULL # usually a call to rstan::sampling()
    stan_object
  }
)

## End(Not run)

jasonmtroos/sbcrs documentation built on Nov. 4, 2019, 2:20 p.m.