Implements an adaptive MCMC algorithm with component-wise updates
and scale adaptation during burn-in.
Usage
AMCMC(
theta_seed,
nsim,
eta = 0.6,
adapt_frequency = 60L,
init_scale = 1,
proposal_dist = "normal"
)
Arguments
- theta_seed
Initial parameter values
- nsim
Total number of MCMC iterations
- eta
Adaptation rate parameter, must be in (0.5, 1)
- adapt_frequency
Frequency of scale adaptations (in iterations)
- init_scale
Initial proposal scale (1/variance)
- proposal_dist
Proposal distribution type, normal or student
Value
Matrix of MCMC samples (rows are iterations, columns are parameters)
Details
The algorithm uses component-wise Metropolis-Hastings updates
with adaptive scaling during the burn-in period (first third of iterations).
The scaling adapts to achieve approximately 36 percent acceptance rate per component.