Predictive Model Weights for brma Objects

Compute stacking or pseudo-BMA weights from LOO results already stored in multiple brma objects.

Usage

# S3 method for class 'brma'
loo_model_weights(
  x,
  ...,
  unit = "estimate",
  method = c("stacking", "pseudobma"),
  optim_method = "BFGS",
  optim_control = list(),
  BB = TRUE,
  BB_n = 1000,
  alpha = 1,
  r_eff_list = NULL,
  cores = getOption("mc.cores", 1)
)

Arguments

x

a brma model object (the first model to weight).

...

additional brma model objects or RoBMA-targeted psis_loo objects.

unit

output/deletion unit used when extracting stored LOO results from brma objects.

method

weighting method; "stacking" or "pseudobma".

optim_method, optim_control

optimization method and controls for stacking. See loo_model_weights.

BB, BB_n, alpha

Bayesian-bootstrap controls for pseudo-BMA weighting. See loo_model_weights.

r_eff_list

optional relative effective sample sizes. Retained for the upstream interface and ignored when stored psis_loo objects are supplied.

cores

number of cores passed to loo::loo_model_weights().

Value

A named numeric vector containing one predictive weight per model.

Details

Each brma object must first be updated with object <- add_loo(object, unit = unit). All models must use the same outcome data, deletion unit, and likelihood target. RoBMA validates these targets before delegating to loo::loo_model_weights().

Model names are derived from the supplied model expressions. See loo_model_weights for details about stacking, pseudo-BMA, and pseudo-BMA+.

See also

add_loo, loo.brma, loo_compare.brma, loo_model_weights

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("metadat", quietly = TRUE)) {
  data(dat.lehmann2018, package = "metadat")

  fit_bias <- RoBMA(yi = yi, vi = vi, data = dat.lehmann2018, measure = "SMD")
  fit_nobias <- BMA(yi = yi, vi = vi, data = dat.lehmann2018, measure = "SMD")

  fit_bias   <- add_loo(fit_bias)
  fit_nobias <- add_loo(fit_nobias)

  loo_model_weights(fit_bias, fit_nobias)
}
} # }