genMCMCDiag (version 0.2.3)

genDiagnostic: Generate Generalized Diagnostics for Markov Chain Monte Carlo Draws

Description

This function generates generalized diagnostics for Markov Chain Monte Carlo (MCMC) draws, transforming the draws if specified, and evaluating selected diagnostics.

Usage

genDiagnostic(
  mhDraws,
  proximityMap = c("standard", "ts", "lanfear"),
  diagnostics = c("traceplot", "ess", "psrf"),
  distance = NULL,
  verbose = FALSE,
  ...
)

Value

An object of class 'mcmcDiag', containing evaluated diagnostics, transformed draws, and function call details.

Arguments

mhDraws

A list of MCMC draws, where each element is an ordered list or numeric vector representing the output of a single MCMC chain.

proximityMap

Method (called a proximity-map) for transforming the MCMC draws. Options include 'standard', 'ts', 'lanfear', or a custom function. See details.

diagnostics

A character vector or list of diagnostic functions to be evaluated. Options include 'traceplot', 'ess', 'psrf', or custom functions. See details.

distance

Function for evaluating distance between MCMC draws if required by 'method'. This should be a pairwise distance function that operates on elements of the chains from mhDraws. Note that the lanfear and ts proximityMaps ALWAYS require a distance function.

verbose

If TRUE, informative messages are displayed.

...

Arguments passed on to tsTransform, lanfearTransform

minDist

Numeric. Value which specifies the minimum possible distance for two draws which are not equal. See tsTransform details.

fuzzy

Logical. If TRUE computes an approximate version of the TS algorithm. See tsTransform details.

fuzzyDist

Numeric. Parameter for approximate version of ts algorithm. See tsTransform details.

reference

Argument for method = 'lanfear'. Reference point for lanfearTransform (with exact same structure as each MCMC draw) for draw comparison. If left NULL a random point is selected from the given draws. See lanfearTransform details.

Details

Built-in proximity-maps can be called with the appropriate character string in the 'proximity-map' argument. For details on a particular proximity-map use ?lanfearTransform or ?tsTransform, the standard proximity-map induces no transformation. Custom proximity-map functions may be added as well. A custom function must be written to accept a list of mcmcChain type objects, and output a list of dataframes with columns val (the transformed draw) and t (the MCMC chain order). Each element in the list is the transformed MCMC chain corresponding to the input.

Built-in diagnostics can be called with the appropriate character string in the 'diagnostics' argument. Current diagnostic options are 'traceplot' for traceplots, 'ess' for Effective Sample Size, and 'psrf' for the Gelman-Rubin Potential Scale Reduction Factor. Additional custom diagnostic functions may be written. These functions should act on a list of data.frames output from a transform function and should output as a relatively small data.frame where the name of diagnostic is the first row.name.

Examples

#Example using standard Traceplot
tstS <- genDiagnostic(uniMCMCResults)
tstS

#Example using 'lanfear' traceplot
tstL <- genDiagnostic(uniMCMCResults, proximityMap = 'lanfear', distance = eucDist,
                      reference = 0)
tstL

#Example using bayesian network sample data, with 'lanfear' proximityMap
tstBN1 <- genDiagnostic(bnMCMCResults, proximityMap = 'lanfear', distance = partitionDist)
tstBN1