Thinning a Markov chain
Topic of Markov chains, Gibbs sampling, and Metropolis–Hastings, part of Bayesian statistics
Thinning a Markov chain
| Type | Topic |
| Taught in | Markov chains, Gibbs sampling, and Metropolis–Hastings |
| Part | Bayesian statistics |
Keeping only every kth draw of an MCMC run. It reduces storage and visible autocorrelation, but it throws information away: the full chain gives a Monte Carlo average at least as precise, so thinning is justified only by storage or by an expensive later calculation.
In Markov chains, Gibbs sampling, and Metropolis–Hastings
Thinning by retaining every kth draw reduces storage and visible serial dependence, but usually throws away usable information. Unless storage or an expensive downstream calculation is the constraint, retain all post-warm-up draws and quantify their effective sample size. Posterior means, probabilities, and equal-tailed intervals are computed from those draws; a continuous MAP is found by optimisation, not by asking which simulated value occurred most often.
Used in
See also
Markov chain · Autocorrelation function · Monte Carlo method · Warm-up is not a convergence proof · Stationarity and detailed balance · Detailed balance implies stationarity · Systematic-scan Gibbs sampler · Metropolis–Hastings update