Skip to contents

Check MCMC Convergence

Source: R/post_methods.R
check_convergence.Rd

Prints effective sample sizes (ESS) and Geweke diagnostics for the scalar variance parameters in a fitted DBN model. Always run this before interpreting results to verify the MCMC sampler has converged.

Usage

check_convergence(results)

Arguments

results

Output from dbn()

Value

Invisible NULL (diagnostics are printed to console)

Details

Effective sample size (ESS): The number of effectively independent posterior draws. Values above 200-400 are generally adequate. Low ESS means the chain is highly autocorrelated and you should increase nscan or odens.

Geweke diagnostic: Tests whether the first and last portions of the chain come from the same distribution. Absolute values above 2 suggest the chain has not converged and you should increase burn.

For visual diagnostics, use plot_trace() to inspect trace plots.

See also

dbn, compare_dbn, param_summary

Examples

# \donttest{
sim <- simulate_static_dbn(n = 6, time = 10, seed = 1)
fit <- dbn(sim$Y, model = "static", nscan = 200, burn = 100, verbose = FALSE)
check_convergence(fit)
#> ℹ Fixed parameters (not sampled): "s2"
#> 
#> ── Effective Sample Sizes 
#>         t2         g2 
#> 246.975262   7.237814 
#> 
#> ── Geweke Diagnostic 
#> 
#> Fraction in 1st window = 0.1
#> Fraction in 2nd window = 0.5 
#> 
#>     t2     g2 
#> -2.115 -5.600 
#> 
# }