EcoEnsemble (version 1.1.2)

sample_prior: Generate samples of latent variables from prior predictive distribution

Description

Methods to generates samples of the latent variables from the prior predictive distribution of the ensemble model.

Usage

sample_prior(
  observations,
  simulators,
  priors,
  sam_priors,
  num_samples = 1,
  full_sample = TRUE,
  drivers = FALSE,
  ...
)

Value

An EnsembleSample object.

Arguments

observations

A list of length 2 containing observations and a covariance matrix. The first element is a data.frame or matrix with each column giving observations of each output of interest and each row a time. Rows should be named with the times and columns should be named the variables. The second element is is a \(d \times d\) matrix where \(d\) is the number of columns of the observations data frame / matrix. This matrix is the covariance matrix of the observations.

simulators

A list with length equal to the number of simulators. For each simulator, there is a list of 2 objects containing the simulator output and covariance matrix. The first element is a data.frame or matrix with each column giving a simulator outputs of interest and each row a time. Rows should be named with the times and columns should be named the variables. The second element is a \(n_k \times n_k\) matrix where \(n_k\) is the number of columns of the simulators output data frame / matrix. This matrix is the covariance matrix of the simulator outputs.

priors

An EnsemblePrior object specifying the prior distributions for the ensemble.

sam_priors

A list containing two items named samples and point_estimate. samples is either a NULL or a stanfit object containing the samples drawn from the prior distribution of the ensemble model and point_estimate is either a NULL or a list object containing the optimised prior distribution of the ensemble model. If this object is missing then sample_prior generates it.

num_samples

A numeric specifying the number of samples to be generated. The default is 1.

full_sample

A logical that runs a full sampling of the prior density of the ensemble model if TRUE. If FALSE, returns the point estimate which maximises the prior density of the ensemble model.

drivers

A logical indicating whether drivers have been used in combination with simulators. Default value is FALSE.

...

Additional arguments passed to the function rstan::sampling or rstan::optimizing.

Details

The samples are created using the methods described in Strickland et. al. (2009) and Durbin and Koopman (2002).

References

J. Durbin, S. J. Koopman (2002) A simple and efficient simulation smoother for state space time series analysis Biometrika, Volume 89, Issue 3, August 2002, Pages 603-616,

Chris M.Strickland, Ian. W.Turner, RobertDenhamb, Kerrie L.Mengersena. Efficient Bayesian estimation of multivariate state space models Computational Statistics & Data Analysis Volume 53, Issue 12, 1 October 2009, Pages 4116-4125

See Also

EnsembleFit, EnsembleSample, generate_sample, prior_ensemble_model

Examples

# \donttest{
priors <- EnsemblePrior(4)
prior_density <- prior_ensemble_model(priors, M = 4)
samples <- sample_prior(observations = list(SSB_obs, Sigma_obs),
             simulators = list(list(SSB_miz, Sigma_miz),
                               list(SSB_ewe, Sigma_ewe),
                               list(SSB_fs, Sigma_fs),
                               list(SSB_lm, Sigma_lm)),
             priors = priors,
             sam_priors = prior_density)
plot(samples) #Plot the prior predictive density.
# }