Bayesian Synthetic Likelihood
Bayesian synthetic likelihood (BSL, Price et al. (2018) <10.1080>)
is an alternative to standard, non-parametric approximate Bayesian
computation (ABC). BSL assumes a multivariate normal distribution
for the summary statistic likelihood and it is suitable when the
distribution of the model summary statistics is sufficiently regular.
This package provides a Metropolis Hastings Markov chain Monte Carlo
implementation of four methods (BSL, uBSL, semiBSL and BSLmisspec) and two
shrinkage estimators (graphical lasso and Warton's estimator).
uBSL (Price et al. (2018) <10.1080>) uses
an unbiased estimator to the normal density. A semi-parametric version
of BSL (semiBSL, An et al. (2018) <1809.05800>) is more robust
to non-normal summary statistics. BSLmisspec (Frazier et al. 2019
<1904.04551>) estimates the Gaussian synthetic likelihood whilst
acknowledging that there may be incompatibility between the model and the
observed summary statistic. Shrinkage estimation can help to decrease the
number of model simulations when the dimension of the summary statistic is
high (e.g., BSLasso, An et al. (2019) <10.1080>).
Extensions to this package are planned.10.1080>1904.04551>1809.05800>10.1080>10.1080>