models.pseudo_bma_weights()

Pseudo-BMA weights, w_k proportional to exp(elpd_k); with

Usage

models.pseudo_bma_weights(
    lpd,
    bootstrap=True,
    n_draws=1000,
    seed=0,
)

bootstrap=True, pseudo-BMA+ weights averaged over Bayesian-bootstrap replicates of the observations, which keeps a model that is only slightly better from taking all the weight.

Parameters

lpd: list of list of float

One list per model, as for stacking_weights.

bootstrap: bool = True

n_draws: int = 1000

Bootstrap replicates.

seed: int = 0

Replicate b uses stream b of seed.

Returns

list of float