Burn-in
Topic of Markov chains, Gibbs sampling, and Metropolis–Hastings, part of Bayesian statistics · Also in Bayesian linear regression
Burn-in
| Type | Topic |
| Taught in | Markov chains, Gibbs sampling, and Metropolis–Hastings · Bayesian linear regression |
| Part | Bayesian statistics |
| Questions using it | 13 |
The initial stretch of a Markov chain Monte Carlo run that is discarded because the chain may not yet have forgotten its starting value; the course also calls it warm-up. Discarding it limits dependence on the start. It does not prove that the rest of the run has converged.
In Markov chains, Gibbs sampling, and Metropolis–Hastings
Discarding early draws is called warm-up or burn-in. It limits dependence on the chosen initial state, but it is not a proof of convergence. Strong posterior correlation produces staircase-like Gibbs paths and high autocorrelation. Separated modes are harder still because a coordinate-wise move may have to cross a low-density barrier.
In Bayesian linear regression
The resulting pairs are iid joint-posterior draws. There is no burn-in, serial dependence, or convergence diagnostic.
Used in
- Taught in Markov chains, Gibbs sampling, and Metropolis–Hastings · Bayesian linear regression
- Also used in Bayesian linear regression
- Questions that use it (13) — Metropolis–Hastings for the logit-normal posterior · Q4 (answer) · Q5 (answer) · Q25 (answer) · Q26 (answer) · Q42 (answer) · Q43 (answer) · Q58 (answer) · Question 3 (answer) · Question 9 (answer) · Question 12 (answer) · Question 16 (answer) · C15
- All 31 questions on Markov chains, Gibbs sampling, and Metropolis–Hastings
See also
Warm-up is not a convergence proof · Direct posterior simulation · Markov chain · Stationarity and detailed balance · Detailed balance implies stationarity · Systematic-scan Gibbs sampler · Metropolis–Hastings update · Metropolis–Hastings invariance