spatialGEV_predict: Draw from the posterior predictive distributions at new...

In SpatialGEV: Fit Spatial Generalized Extreme Value Models

Draw from the posterior predictive distributions at new locations based on a fitted GEV-GP model

Description

Draw from the posterior predictive distributions at new locations based on a fitted GEV-GP model

Usage

spatialGEV_predict(
  model,
  locs_new,
  n_draw,
  type = "response",
  X_a_new = NULL,
  X_b_new = NULL,
  X_s_new = NULL,
  parameter_draws = NULL
)

Arguments

model	A fitted spatial GEV model object of class spatialGEVfit
locs_new	A ⁠n_test x 2⁠ matrix containing the coordinates of the new locations
n_draw	Number of draws from the posterior predictive distribution
type	A character string: "response" or "parameters". The former returns draws from the posterior predictive distribution, and the latter returns parameter draws (all on original scale).
X_a_new	⁠n_test x r1⁠ design matrix for a at the new locations. If not provided, the default is a column matrix of all 1s.
X_b_new	⁠n_test x r2⁠ design matrix for log(b) at the new locations
X_s_new	⁠n_test x r2⁠ design matrix for (possibly transformed) s at the new locations
parameter_draws	Optional. A ⁠n_draw x n_parameter⁠ matrix, or an object that is of class 'spatialGEVsam'. If spatialGEV_sample() has already been called, the output matrix of parameter draws can be supplied here to avoid doing sampling of parameters again. Make sure the number of rows of parameter_draws is the same as n_draw.

Value

An object of class spatialGEVpred, which is a list of the following components:

• An ⁠n_draw x n_test⁠ matrix pred_y_draws containing the draws from the posterior predictive distributions at n_test new locations

• An ⁠n_test x 2⁠ matrix locs_new containing the coordinates of the test data

• An ⁠n_train x 2⁠ matrix locs_obs containing the coordinates of the observed data

Examples

set.seed(123)
library(SpatialGEV)
n_loc <- 20
a <- simulatedData$a[1:n_loc]
logb <- simulatedData$logb[1:n_loc]
logs <- simulatedData$logs[1:n_loc]
y <- simulatedData$y[1:n_loc]
locs <- simulatedData$locs[1:n_loc,]
n_test <- 5
test_ind <- sample(1:n_loc, n_test)

# Obtain coordinate matrices and data lists
locs_test <- locs[test_ind,]
y_test <- y[test_ind]
locs_train <- locs[-test_ind,]
y_train <- y[-test_ind]

# Fit the GEV-GP model to the training set
train_fit <- spatialGEV_fit(
  data = y_train,
  locs = locs_train,
  random = "ab",
  init_param = list(
    beta_a = mean(a),
    beta_b = mean(logb),
    a = rep(0, n_loc-n_test),
    log_b = rep(0, n_loc-n_test),
    s = 0,
    log_sigma_a = 1,
    log_kappa_a = -2,
    log_sigma_b = 1,
    log_kappa_b = -2
  ),
  reparam_s = "positive",
  kernel = "matern",
  silent = TRUE
)

pred <- spatialGEV_predict(
  model = train_fit,
  locs_new = locs_test,
  n_draw = 100
)
summary(pred)

SpatialGEV documentation built on June 22, 2024, 9:24 a.m.