Getting Started with GWPR.light 1.0.0

The hardware and bandwidth for this mirror is donated by dogado GmbH, the Webhosting and Full Service-Cloud Provider. Check out our Wordpress Tutorial.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]dogado.de.

Overview

GWPR.light 1.0.0 provides a modern, sf-first API for Geographically Weighted Panel Regression (GWPR). The public interface consists of four functions:

gwpr() — full pipeline (bandwidth search + fitting + diagnostics)
select_bandwidth() — standalone bandwidth optimisation
fit_gwpr() — fit with a known bandwidth
diagnose_gwpr() — diagnostic tests on a fitted model

All functions accept panel data as a plain data.frame and spatial information as an sf object. The workers argument controls parallel execution; the default workers = 1 runs serially and is safe in all environments.

Minimal linear GWPR example

library(GWPR.light)
library(sf)
#> Warning: package 'sf' was built under R version 4.3.3
#> Linking to GEOS 3.13.0, GDAL 3.8.5, PROJ 9.5.1; sf_use_s2() is TRUE

set.seed(42)

# Simulate a tiny spatial panel: 6 units, 4 time periods
n_units <- 6
n_time  <- 4

pts <- sf::st_as_sf(
  data.frame(
    id = 1:n_units,
    X  = c(0, 1, 2, 0, 1, 2),
    Y  = c(0, 0, 0, 1, 1, 1)
  ),
  coords = c("X", "Y"),
  crs    = NA_integer_
)

dat <- data.frame(
  id   = rep(1:n_units, each = n_time),
  time = rep(1:n_time,  n_units),
  x1   = rnorm(n_units * n_time),
  x2   = rnorm(n_units * n_time)
)
dat$y <- 1.5 * dat$x1 - 0.8 * dat$x2 + rnorm(n_units * n_time, sd = 0.3)

# Fit with a known bandwidth (skip automatic search for speed)
fit <- fit_gwpr(
  formula   = y ~ x1 + x2,
  data      = dat,
  spatial   = pts,
  id        = "id",
  time      = "time",
  bandwidth = 2,
  family    = "gaussian",
  model     = "pooling",
  workers   = 1
)

print(fit)
#> Geographically Weighted Panel Regression (gwpr_fit)
#> ----------------------------------------------------
#> Family   : gaussian 
#> Model    : pooling 
#> Effect   : individual 
#> Bandwidth: 2 
#> Metrics  :
#>   R2           0.9816
#>   MSE          0.06987
#>   RMSE         0.2643
#>   MAE          0.1924

Accessing results

# Overall goodness-of-fit metrics
str(fit$metrics)
#> List of 4
#>  $ R2  : num 0.982
#>  $ MSE : num 0.0699
#>  $ RMSE: num 0.264
#>  $ MAE : num 0.192

# Per-unit spatial coefficients (one row per spatial unit)
if (!is.null(fit$spatial_results)) {
  head(fit$spatial_results)
}
#>   unit_id status coef_(Intercept)  coef_x1    coef_x2
#> 1       1     ok       0.09506111 1.454001 -0.7834870
#> 2       2     ok       0.08224925 1.448562 -0.8178750
#> 3       3     ok       0.03704843 1.441328 -0.8517107
#> 4       4     ok       0.10574378 1.497491 -0.8031366
#> 5       5     ok       0.08602491 1.468562 -0.8307433
#> 6       6     ok       0.03501923 1.440875 -0.8499490

Bandwidth selection (grid search)

bw <- select_bandwidth(
  formula  = y ~ x1 + x2,
  data     = dat,
  spatial  = pts,
  id       = "id",
  time     = "time",
  family   = "gaussian",
  model    = "pooling",
  method   = "grid",
  control  = list(lower = 0.5, upper = 3, step = 0.5),
  workers  = 1
)

print(bw)
#> Bandwidth Search Result (gwpr_bandwidth)
#> ----------------------------------------
#> Method        : grid 
#> Criterion     : MSE 
#> Best bandwidth: 0.5 
#> Best score    : 0.04480074 
#> Iterations    : 6 
#> Elapsed (s)   : 0.03
cat("Best bandwidth:", bw$best_bandwidth, "\n")
#> Best bandwidth: 0.5

Full pipeline with gwpr()

# Use the best bandwidth found above to avoid re-running search
full_fit <- gwpr(
  formula     = y ~ x1 + x2,
  data        = dat,
  spatial     = pts,
  id          = "id",
  time        = "time",
  bandwidth   = bw$best_bandwidth,
  family      = "gaussian",
  model       = "pooling",
  diagnostics = FALSE,   # skip diagnostics for speed
  workers     = 1
)

print(full_fit)
#> Geographically Weighted Panel Regression (gwpr_fit)
#> ----------------------------------------------------
#> Family   : gaussian 
#> Model    : pooling 
#> Effect   : individual 
#> Bandwidth: 0.5 
#> Metrics  :
#>   R2           0.9882
#>   MSE          0.0448
#>   RMSE         0.2117
#>   MAE          0.1425

Diagnostics

diag_result <- diagnose_gwpr(
  full_fit,
  diagnostics = c("f_test", "hausman", "lm_test")
)

print(diag_result)
#> GWPR Diagnostics (gwpr_diagnostics)
#> ------------------------------------
#> Model type   : gaussian 
#> Panel balance: TRUE 
#> Tests run    : f_test, hausman, lm_test

Long-running examples

The following code illustrates automatic bandwidth search via SGD and a binomial (logistic) GWPR. These are wrapped in \donttest{} in the function documentation because they may take more than a few seconds on larger datasets.

# Automatic SGD bandwidth search + fit (may take several seconds)
fit_auto <- gwpr(
  formula          = y ~ x1 + x2,
  data             = dat,
  spatial          = pts,
  id               = "id",
  time             = "time",
  bandwidth_method = "sgd",
  bandwidth_control = list(n_iter = 20, step_size = 0.1),
  workers          = 1,
  seed             = 123
)

# Binomial GWPR
dat$y_bin <- as.integer(dat$y > 0)
fit_logit <- fit_gwpr(
  formula   = y_bin ~ x1 + x2,
  data      = dat,
  spatial   = pts,
  id        = "id",
  time      = "time",
  bandwidth = 2,
  family    = "binomial",
  model     = "pooling",
  workers   = 1
)

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.
Health stats visible at Monitor.