laplace_approx
Calculate model evidence using the Laplace approximation
Specification
Alias: None
Arguments: None
Description
The laplace_approx keyword for model evidence indicates that
a pre-solve will be used prior to the Bayesian MCMC sampling
to estimate the Maximum A Posteriori (MAP) point. The Laplace
approximation assumes the posterior density is nearly
Gaussian and is given by a formula which involves the likelihood
at the MAP point, the prior density at the MAP point, and the
Hessian of the log-posterior at the MAP point. The formula is
given in the Dakota User’s manual. This method is efficient
at estimating the model evidence for posterior densities with weak
non-Gaussian characteristics but it does require a MAP
solve (so pre-solve should be specified) and it does
require gradient and Hessians of the response to be on.
Default Behavior
Expected Output Currently, the model evidence will be printed in the screen output with prefacing text indicating if it is calculated by the Laplace approximation.
Usage Tips
