Skip to contents

Prepare a new nested sampling run

Source: R/ernest_sampler.R
ernest_sampler.Rd

Initializes an ernest_sampler object containing the components required to perform nested sampling. This object can then be used to build sequences of nested samples with generate().

Usage

ernest_sampler(
  log_lik,
  prior,
  sampler = rwmh_cube(),
  n_points = 500,
  first_update = NULL,
  update_interval = NULL
)

Arguments

log_lik

A function which takes a vector of parameters as a vector and returns the log-likelihood of a given model. Wrapped using create_likelihood() unless it is already of class ernest_likelihood.

prior

An object with class ernest_prior. Describes the prior space within which to generate sample parameters for log_lik.

sampler

An object of class ernest_lrps. Describes the likelihood-restricted prior sampling technique to adopt during the run.

n_points

A strictly positive integer. The number of live points to use in the nested sampling run.

first_update

An optional positive integer. The number of likelihood calls to make with the default uniform LRPS method before swapping to the technique described by sampler. If left NULL, this is set to n_points * 2.5.

update_interval

An optional positive integer. The number of likelihood calls between updates to the sampler object. If NULL, this is set to n_points * 1.5.

Value

An object of class ernest_sampler, which is a list containing the inputs used as arguments to this function, along with an environment run_env which is used to store the n_points live particles throughout a nested sampling run.

Details

The ernest_sampler object is tested with compile() before it is returned. This helps to catch errors with the likelihood and prior specifications. If this compilation step fails, review your log_lik_fn and prior objects for their compliance.

Verbosity

Messages from ernest can be silenced with the global options rlib_message_verbosity and rlib_warning_verbosity. These options take the values:

  • "default": Verbose unless the .frequency argument is supplied.

  • "verbose": Always verbose.

  • "quiet": Always quiet.

When set to quiet, message are not displayed and the condition is not signaled. See rlang::abort() for more information.

See also

  • create_likelihood() describes the requirements of the log_lik_fn parameter.

  • create_prior() describes the requriements of the prior parameter.

  • ernest_lrps describes the general requirements of likelihood-restricted

  • prior samplers.

Examples

prior <- create_uniform_prior(n_dim = 2, lower = -1, upper = 1)
#> New names:
#> • `Uniform` -> `Uniform...1`
#> • `Uniform` -> `Uniform...2`
ll_fn <- function(x) -sum(x^2)
sampler <- ernest_sampler(ll_fn, prior, n_points = 100)
sampler
#> Nested sampling specification <ernest_sampler>
#> No. Points: 100
#> 
#> ── Sampling Method 
#> • Random Walk in Unit Cube LRPS <rwmh_cube/ernest_lrps>
#> • No. Dimensions: 2
#> • No. Calls Since Update: 0
#> • No. Accepted Since Update: 0
#> • Current Step Size: 1

# Use a unit-cube LRPS (not recommended in practice)
unit_sampler <- ernest_sampler(
  ll_fn,
  prior,
  n_points = 100,
  sampler = unif_cube()
)
unit_sampler
#> Nested sampling specification <ernest_sampler>
#> No. Points: 100
#> 
#> ── Sampling Method 
#> • ! An abstract LRPS sampler <ernest_lrps>