Predict from a fitted geographically weighted model

For Ridge, Lasso, Elastic Net, SCAD and MCP, new target locations are fitted from retained training data using only their training neighbors. `newdata` supplies predictor values at targets, while `newcoords` controls the local training samples.

Usage

# S3 method for class 'gwrs_fit'
predict(
  object,
  newdata = NULL,
  newcoords = NULL,
  neighbors = NULL,
  k = NULL,
  type = c("response", "coefficients", "both"),
  ...
)

Arguments

object

A fitted `gwrs_fit`.

newdata

Optional target design matrix or formula data frame.

newcoords

Coordinates corresponding to `newdata`.

neighbors

Optional cross-neighbor object constructed with `gwr_neighbors(training_coords, query_coords = newcoords, ...)`.

k

Number of training neighbors when `neighbors` is not supplied.

type

Return predictions, local coefficients, or both.

...

Unused.

Value

A numeric prediction vector, coefficient matrix, or a list of both.

Examples

train <- datasets::quakes[1:30, ]
target <- datasets::quakes[31:33, ]
train_coords <- as.matrix(train[, c("long", "lat")])
train_x <- as.matrix(train[, c("depth", "stations")])
fit <- gwr_en_fit(
  train_x, train$mag, coords = train_coords, k = 18,
  lambda = 0.05, alpha = 0.7,
  control = gwrs_control(n_threads = 1)
)
prediction <- predict(
  fit,
  newdata = as.matrix(target[, c("depth", "stations")]),
  newcoords = as.matrix(target[, c("long", "lat")]),
  k = 18,
  type = "both"
)
prediction$fit
prediction$coefficients