ensemble: Simulate an ensemble of posterior predictions

In joepowers16/rethinking: Statistical Rethinking book package

Description

Uses link and sim for a list of map or map2stan model fits to construct Akaike weighted ensemble of predictions.

Usage

ensemble( ... , data , n=1e3 , func=WAIC , weights , WAIC=TRUE , refresh=0 , 
          replace=list() , do_link=TRUE , do_sim=TRUE )

Arguments

...	map or map2stan models
data	Optional data to compute predictions over, as in link and sim
n	Number of samples to draw from posterior for each model
func	Function to use in computing criterion for model weights
weights	Optional vector of weights to use. If present, func is ignored.
WAIC	Deprecated: If TRUE, use func to compute weights. Otherwise tries to use DIC.
refresh	Progress update refresh interval. 0 suppresses output.
replace	Optional named list with replacement posterior samples. Used for maginalizing over random effects, for example. See example in link.
do_link	If TRUE, compute and return link results
do_sim	If TRUE, compute and return sim results

Details

This function calls link and sim for each fit model given as input. The results are then combined into ensemble link and simulation output, where samples from each model are represented in proportion to the Akaike weights. Akaike weights are calculated by compare, using func, unless an explicit vector weights is provided. The values in weights will be normalized to sum to one, if they do not already.

Note that no averaging is done by this function. Parameters are not averaged, and predictions are not averaged.

Author(s)

Richard McElreath

See Also

link, sim, compare

joepowers16/rethinking documentation built on June 2, 2019, 6:52 p.m.