Dakota Reference Manual
Version 6.4
LargeScale Engineering Optimization and Uncertainty Analysis

(Experimental) Gaussian Process Models for Simulation Analysis (GPMSA) Markov Chain Monte Carlo algorithm with Gaussian Process Surrogate
This keyword is related to the topics:
Alias: none
Argument(s): none
Required/Optional  Description of Group  Dakota Keyword  Dakota Keyword Description  

Required  chain_samples  Number of Markov Chain Monte Carlo posterior samples  
Optional  seed  Seed of the random number generator  
Required  build_samples  Number of initial model evaluations used in build phase  
Optional  import_build_points_file  File containing points you wish to use to build a surrogate  
Optional (Choose One)  MCMC algorithm type (Group 1)  dram  Use the DRAM MCMC algorithm  
delayed_rejection  Use the Delayed Rejection MCMC algorithm  
adaptive_metropolis  Use the Adaptive Metropolis MCMC algorithm  
metropolis_hastings  Use the MetropolisHastings MCMC algorithm  
multilevel  Use the multilevel MCMC algorithm.  
Optional  rng  Selection of a random number generator  
Optional  pre_solve  Perform deterministic optimization for MAP before Bayesian calibration  
Optional  proposal_covariance  Defines the technique used to generate the MCMC proposal covariance. 
GPMSA (Gaussian Process Models for Simulation Analysis) is another approach that provides the capability for Bayesian calibration. The GPMSA implementation currently is an experimental capability and not ready for production use at this time. A key part of GPMSA is the construction of an emulator from simulation runs collected at various settings of input parameters. The emulator is a statistical model of the system response, and it is used to incorporate the observational data to improve system predictions and constrain or calibrate the unknown parameters. The GPMSA code draws heavily on the theory developed in the seminal Bayesian calibration paper by Kennedy and O'Hagan[54]. The particular approach in GPMSA has been developed by the Los Alamos group and document in[47]. GPMSA uses Gaussian process models in the emulation, but the emulator is actually a set of basis functions (e.g. from a singular value decomposition) which have GPs as the coefficients.
For the GPMSA method, one can define the number of samples which will be used in construction of the emulator, build_samples
. The emulator involves Gaussian processes in GPMSA, so the user does not specify anything about emulator type. At this point, the only controls active for GPMSA are build_samples
, seed
and rng
, and samples
(the number of MCMC samples) and the type of MCMC algorithm (e.g. dram
, delayed_rejection
, adaptive_metropolis
, metropolis_hastings
, or multilevel
). NOTE: the GPMSA method is in a very preliminary, prototype state at this time. The user will need to modify certain data structures in the code for their particular application and recompile to run with GPMSA.