coda
Output analysis and diagnostics for MCMC.
MCMC Objects
library(coda)
# Create mcmc object
mcmc_obj <- mcmc(matrix_data)
# With parameters
mcmc_obj <- mcmc(matrix_data,
start = 1,
end = 10000,
thin = 1
)
# Multiple chains
chains <- mcmc.list(chain1, chain2, chain3)
Summary Statistics
# Summary
summary(mcmc_obj)
# Mean
colMeans(mcmc_obj)
# Standard deviation
apply(mcmc_obj, 2, sd)
# Quantiles
summary(mcmc_obj)$quantiles
Convergence Diagnostics
# Geweke diagnostic
geweke.diag(mcmc_obj)
geweke.plot(mcmc_obj)
# Heidelberg-Welch diagnostic
heidel.diag(mcmc_obj)
# Raftery-Lewis diagnostic
raftery.diag(mcmc_obj)
# Gelman-Rubin (requires multiple chains)
gelman.diag(chains)
gelman.plot(chains)
Effective Sample Size
# Effective sample size
effectiveSize(mcmc_obj)
# Autocorrelation
autocorr(mcmc_obj)
autocorr.diag(mcmc_obj)
autocorr.plot(mcmc_obj)
Plotting
# Trace plot
traceplot(mcmc_obj)
plot(mcmc_obj, trace = TRUE, density = FALSE)
# Density plot
densplot(mcmc_obj)
plot(mcmc_obj, trace = FALSE, density = TRUE)
# Both
plot(mcmc_obj)
# Autocorrelation plot
autocorr.plot(mcmc_obj)
# Cross-correlation
crosscorr.plot(mcmc_obj)
Credible Intervals
# HPD interval
HPDinterval(mcmc_obj)
HPDinterval(mcmc_obj, prob = 0.95)
# Quantile-based interval
summary(mcmc_obj)$quantiles[, c("2.5%", "97.5%")]
Thinning and Subsetting
# Thin chain
thinned <- window(mcmc_obj, thin = 10)
# Subset iterations
subset <- window(mcmc_obj, start = 1000, end = 5000)
# Subset parameters
subset <- mcmc_obj[, c("beta1", "beta2")]
Combining Chains
# Combine multiple chains
combined <- mcmc.list(chain1, chain2, chain3)
# Stack chains
stacked <- as.mcmc(do.call(rbind, combined))
Cross-Correlation
# Cross-correlation matrix
crosscorr(mcmc_obj)
# Plot
crosscorr.plot(mcmc_obj)
Batch Means
# Batch means standard error
batchSE(mcmc_obj)
# Spectrum at zero
spectrum0(mcmc_obj)
spectrum0.ar(mcmc_obj)
Reading/Writing
# Read CODA output
mcmc_obj <- read.coda("chain.txt", "index.txt")
# Write
write.coda(mcmc_obj, "output")
With Other Packages
# From MCMCpack
library(MCMCpack)
model <- MCMCregress(y ~ x, data = df)
mcmc_obj <- as.mcmc(model)
# From rstan
library(rstan)
mcmc_obj <- As.mcmc.list(stanfit)
# From brms
library(brms)
mcmc_obj <- as.mcmc(brms_fit)
Diagnostic Summary
# Run all diagnostics
diagnostic_summary <- function(mcmc_obj) {
list(
summary = summary(mcmc_obj),
geweke = geweke.diag(mcmc_obj),
heidel = heidel.diag(mcmc_obj),
raftery = raftery.diag(mcmc_obj),
ess = effectiveSize(mcmc_obj)
)
}