PlotGelmanRubinDiagnosticSingleChain {carbondate} | R Documentation |
Plot Histogram of the Gelman-Rubin Convergence Diagnostic for a Single MCMC Chain
Description
This plots a histogram of the potential scale reduction factors (PSRF) for each of the individual posterior
calendar age estimates within a single MCMC chain. Achieved by splitting the chain into segments
after n_burn
and comparing the within-chain variance with the between-chains
variance of the segments. The PSRF of each sample's posterior calendar age is calculated.
If the chain have converged to the target posterior distribution, then PSRF should be close to 1
for all of the samples (a stringent condition is that all values are less than 1.1).
For more information read the vignette:
vignette("determining-convergence", package = "carbondate")
Usage
PlotGelmanRubinDiagnosticSingleChain(output_data, n_burn = NA, n_segments = 3)
Arguments
output_data |
The return value from one of the updating functions, e.g., PolyaUrnBivarDirichlet, WalkerBivarDirichlet or PPcalibrate. |
n_burn |
The number of MCMC iterations that should be discarded for burn-in. This relates to
the total number of iterations |
n_segments |
The number of segments to split the chain into. Default is 3, must be a number between 2 and 10. |
Value
None
Examples
# Plot results for the example data - n_iter is too small for convergence
# Try increasing n_iter to see the values of the PSRF decrease
polya_urn_output <- PolyaUrnBivarDirichlet(
two_normals$c14_age,
two_normals$c14_sig,
intcal20,
n_iter = 500,
show_progress = FALSE)
PlotGelmanRubinDiagnosticSingleChain(polya_urn_output)