Bayesian Hierarchical Active Learning (Pilar 7 x Pilar 5)

Ranks candidate sites by their predictive variance averaged over the BHS posterior MCMC draws – Thompson-sampling-style AL in the language of Settles (2009, Section 3.3).

Usage

al_query_bhs(bhs_fit, pool_data, n_select = 5L, n_draws = 200L)

Arguments

bhs_fit

An edaphos_bhs from bhs_fit().

pool_data

Data frame of candidate sites with covariates + coords.

n_select

Integer; number of candidates to return.

n_draws

Integer; number of posterior draws to average.

Value

edaphos_al_bhs_query data frame sorted by descending average predictive variance.