EMC2 (version 3.5.0)

add_ICs_MLL: Add information criteria to emc object

Description

Adds DIC, BPIC, and optionally MLL values as attributes to an emc object. Can be useful to offload computational burden.

Usage

add_ICs_MLL(
  emc,
  stage = "sample",
  filter = NULL,
  use_best_fit = TRUE,
  BayesFactor = TRUE,
  cores_for_props = 4,
  cores_per_prop = 1,
  ...
)

Value

An emc object with new attributes 'ICs' and 'MLL'

Arguments

emc

List of samples objects

stage

A string. Specifies which stage the samples are to be taken from "preburn", "burn", "adapt", or "sample"

filter

An integer or vector. If it's an integer, iterations up until the value set by filter will be excluded. If a vector is supplied, only the iterations in the vector will be considered.

use_best_fit

Boolean, defaults to TRUE, uses the minimal or mean likelihood (whichever is better) in the calculation, otherwise always uses the mean likelihood.

BayesFactor

Boolean, defaults to TRUE. Include marginal likelihoods as estimated using WARP-III bridge sampling. Usually takes a minute per model added to calculate

cores_for_props

Integer, how many cores to use for the Bayes factor calculation, here 4 is the default for the 4 different proposal densities to evaluate, only 1, 2 and 4 are sensible.

cores_per_prop

Integer, how many cores to use for the Bayes factor calculation if you have more than 4 cores available. Cores used will be cores_for_props * cores_per_prop. Best to prioritize cores_for_props being 4 or 2

...

Additional, optional arguments

Examples

# \donttest{
samples_with_ICs <- add_ICs_MLL(samples_LNR, cores_for_props = 1)
attr(samples_with_ICs, 'MLL')
attr(samples_with_ICs, 'ICs')

# Pre-computed MLLs and ICs are extracted when using compare():
compare(sList=list(samples_with_ICs, samples_LNR), cores_for_props=1)
# Returns the same MD (barring noise), BPIC, DIC for both emc objects, as expected -
# but the first is extracted, the second computed in compare().
# }