Componentwise adaptation for high dimensional MCMC (Q2488398): Difference between revisions
From MaRDI portal
Latest revision as of 09:33, 30 July 2024
scientific article
| Language | Label | Description | Also known as |
|---|---|---|---|
| English | Componentwise adaptation for high dimensional MCMC |
scientific article |
Statements
Componentwise adaptation for high dimensional MCMC (English)
0 references
24 May 2006
0 references
A single component adaptive Metropolis (SCAM) sampling algorithm is described in which the variance of (Gaussian) proposal one-dimensional distribution varies at each step depending on the variancies of previously generated sample. The target components of the SCAM suffer from the poor mixing distribution when the target is highly correlated. So it is recommended to rotate the proposal distribution to the (estimated) principal components of the target distribution. Simulation examples with Gaussian targets in various dimensions up to 1000 are considered.
0 references
principal components
0 references
multivariate distribution
0 references
random vectors generating
0 references
Markov chain Monte Carlo (MCMC) method
0 references
numerical examples
0 references
single component adaptive Metropolis sampling algorithm
0 references
