elpd {BayesSUR}

expected log pointwise predictive density

Description

Measure the prediction accuracy by the elpd (expected log pointwise predictive density). The out-of-sample predictive fit can either be estimated by Bayesian leave-one-out cross-validation (LOO) or by widely applicable information criterion (WAIC) (Vehtari et al. 2017).

Usage

elpd(object, method = "LOO")

Arguments

object

an object of class BayesSUR

method

the name of the prediction accuracy index. Default is the "LOO" (Bayesian LOO estimate of out-of-sample predictive fit). The other index is the "WAIC" (widely applicable information criterion). For the HRR models, both "LOO" and "WAIC" are computed based on the multivate t-distribution of the posterior predictive rather than approximation of importance sampling.

Value

Return the predictiion accuracy measure from an object of class BayesSUR. It is elpd.loo if the argumnet method="LOO" and elpd.WAIC if method="WAIC".

References

Vehtari, A., Gelman, A., Gabry, J. (2017). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Statistics and Computing, 27(5): 1413–1432.

Examples

data("exampleEQTL", package = "BayesSUR")
hyperpar <- list(a_w = 2, b_w = 5)

set.seed(9173)
fit <- BayesSUR(
  Y = exampleEQTL[["blockList"]][[1]],
  X = exampleEQTL[["blockList"]][[2]],
  data = exampleEQTL[["data"]], outFilePath = tempdir(),
  nIter = 10, burnin = 0, nChains = 1, gammaPrior = "hotspot",
  hyperpar = hyperpar, tmpFolder = "tmp/", output_CPO = TRUE
)

## check output
# print prediction accuracy elpd (expected log pointwise predictive density)
# by the Bayesian LOO estimate of out-of-sample predictive fit
elpd(fit, method = "LOO")