Skip to contents

Fit a Bayesian hierarchical spatial linear model (Pilar 7)

Source: R/pilar7_bayesian_hierarchical.R
bhs_fit.Rd

The v2.3.0 activation of the Pilar 7 scaffold. Fits a Bayesian spatial linear model y_i = x_i' beta + w_i + eps_i with a latent exponential-correlation Gaussian process on the residual field and returns full posterior draws.

Usage

bhs_fit(
  data,
  formula,
  coords = c("lon", "lat"),
  backend = c("gibbs", "rcpp", "spBayes"),
  nmcmc = 2000L,
  burn = NULL,
  thin = 1L,
  prior_var_beta = 1000,
  prior_ig_a = 2,
  prior_ig_b = 1,
  phi_range = c(0.01, 10),
  seed = NULL,
  verbose = FALSE
)

Arguments

data

A data frame with the response + covariates + spatial coordinates.

formula

A response ~ covariates formula.

coords

Character length-2 (default c("lon", "lat")) giving the coordinate columns.

backend

One of "gibbs" (pure-R Gibbs sampler, default, no external deps), "spBayes" (dispatches to spBayes::spLM when available).

nmcmc

Integer; number of MCMC iterations. Default 2000L.

burn

Integer; burn-in to discard. Default nmcmc %/% 2.

thin

Integer; keep every thin-th post-burn draw. Default 1L.

prior_var_beta

Numeric; Gaussian prior variance on beta.

prior_ig_a, prior_ig_b

Shape and scale of the inverse-Gamma priors on sigma^2 and tau^2.

phi_range

Numeric length-2; bracket for the profile-MLE of the GP rate parameter phi. Default c(0.01, 10).

seed

Optional RNG seed.

verbose

Logical.

Value

An edaphos_bhs S3 object.