delayed_rejection

Use the Delayed Rejection MCMC algorithm

Topics

bayesian_calibration

Specification

  • Alias: None

  • Arguments: None

  • Default: dram

Child Keywords:

Required/Optional

Description of Group

Dakota Keyword

Dakota Keyword Description

Optional

num_stages

Number of stages

Optional

scale_type

Type of scaling to use

Optional

delay_scale

Scaling parameter

Description

This keyword specifies the use of the Delayed Rejection algorithm in which there can be a delay in rejecting samples from the chain. That is, the “DR” part of DRAM is used but the “AM” part is not, rather a regular Metropolis-Hastings algorithm is used.

Default Behavior

Five MCMC algorithm variants are supported: dram, delayed_rejection, adaptive_metropolis, metropolis_hastings, and multilevel. The default is dram.

Usage Tips

If the user knows something about the proposal covariance or the proposal covariance is informed through derivative information, using delayed_rejection is preferred over dram: the proposal covariance is already being informed by derivative information and the adaptive metropolis is not necessary.

Examples

method,
        bayes_calibration queso
          delayed_rejection
          samples = 10000 seed = 348