get_pars: Filter/Manipulate Parameters from emc Object

Description

Underlying function used in most plotting and object handling functions in EMC2. Can for example be used to filter/thin a parameter type (i.e, group-level means mu) and convert to an mcmc.list.

Usage

get_pars(
  emc,
  selection = "mu",
  stage = get_last_stage(emc),
  thin = 1,
  filter = 0,
  map = FALSE,
  add_recalculated = FALSE,
  length.out = NULL,
  by_subject = FALSE,
  return_mcmc = TRUE,
  merge_chains = FALSE,
  subject = NULL,
  flatten = FALSE,
  remove_dup = FALSE,
  remove_constants = TRUE,
  use_par = NULL,
  type = NULL,
  true_pars = NULL,
  chain = NULL,
  covariates = NULL
)

Value

An mcmc.list object of the selected parameter types with the specified manipulations

Arguments

emc: an emc object.
selection: A Character string. Indicates which parameter type to select (e.g., alpha, mu, sigma2, correlation).
stage: A character string. Indicates from which sampling stage(s) to take the samples from (i.e. preburn, burn, adapt, sample)
thin: An integer. By how much to thin the chains
filter: Integer or numeric vector. If an integer is supplied, iterations up until that integer are removed. If a vector is supplied, the iterations within the range are kept.
map: Boolean. If TRUE parameters will be mapped back to the cells of the experimental design using the design matrices. Otherwise the sampled parameters are returned. Only works for selection = mu or selection = alpha.
add_recalculated: Boolean. If TRUE will also add recalculated parameters, such as b in the LBA (b = B + A; see ?LBA), or z in the DDM z = Z*A (see ?DDM) only works when map = TRUE
length.out: Integer. Alternatively to thinning, you can also select a desired length of the MCMC chains, which will be thinned appropriately.
by_subject: Boolean. If TRUE for selections that include subject parameters (e.g. alpha), plot/stats are organized by subject, otherwise by parameter.
return_mcmc: Boolean. If TRUE returns an mcmc.list object, otherwise a matrix/array with the parameter type.
merge_chains: Boolean. If TRUE returns parameter type merged across chains.
subject: Integer (vector) or character (vector). If an integer will select the 'x'th subject(s), if a character it should match subject names in the data which will be selected.
flatten: Boolean. If FALSE for 3-dimensional samples (e.g., correlations: n-pars x n-pars x iterations). organizes by the dimension containing parameter names, otherwise collapses names across the first and second dimension. Does not apply for selection = "alpha"
remove_dup: Boolean. If TRUE removes duplicate values from the samples. Automatically set to TRUE if flatten = TRUE
remove_constants: Boolean. If TRUE removes constant values from the samples (e.g. 0s in the covariance matrix).
use_par: Character (vector). If specified, only these parameters are returned. Should match the parameter names (i.e. these are collapsed when flatten = TRUE and use_par should also be collapsed names).
type: Character indicating the group-level model selected. Only necessary if sampler isn't specified.
true_pars: Set of true_parameters can be specified to apply flatten or use_par on a set of true parameters
chain: Integer. Which of the chain(s) to return
covariates: Only needed with plot for priors and covariates in the design

Examples

Run this code

# E.g. get the group-level mean parameters mapped back to the design
get_pars(samples_LNR, stage = "sample", map = TRUE, selection = "mu")

# Or return the flattened correlation, with 10 iterations per chain
get_pars(samples_LNR, stage = "sample", selection = "correlation", flatten = TRUE, length.out = 10)

Run the code above in your browser using DataLab