adaptive_metropolis

Use the Adaptive Metropolis 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

period_num_steps

Number of steps between updates of the proposal covariance

Optional

starting_step

Number of steps prior to start of proposal covariance adaptation

Optional

adapt_scale

Sample covariance scaling used to define proposal covariance

Description

This keyword specifies the use of the Adaptive Metropolis algorithm. That is, the “AM” part of DRAM is used but the “DR” part is not: specifying this keyword activates only the Adaptive Metropolis part of the MCMC algorithm, in which the covariance of the proposal density is updated adaptively.

Default Behavior

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

Four MCMC variants are currently supported in MUQ: dram, delayed_rejection, metropolis_hastings and adaptive_metropolis.

Usage Tips

If the user knows very little about the proposal covariance, but doesn’t want to incur the cost of using full dram with both delayed rejection and adaptive metropolis, specifying only adaptive_metropolis offers a good strategy.

Examples

method,
        bayes_calibration queso
          adaptive_metropolis
          samples = 10000 seed = 348