Skip to contents

MLSampling Quickstart Workflow

Source: vignettes/ml-sampling-quickstart.Rmd
ml-sampling-quickstart.Rmd

Overview

This vignette walks through a minimal end-to-end workflow using MLSampling. The example creates synthetic spatial inputs, runs both BDL (Bayesian Deep Learning) and RF (Random Forest) optimizations, and exports results for reporting.

To execute the code, ensure the suggested dependencies listed in DESCRIPTION (such as sf, terra, and torch) are installed.

Load the package

Prepare synthetic field data

The tool expects a boundary geometry, associated covariate rasters, and optional existing sampling locations. The helpers below construct these objects using trivial synthetic data so the example can run without external files.

# Create a simple square boundary in EPSG:32633 (UTM 33N)
boundary <- st_as_sf(st_sfc(st_polygon(list(rbind(
  c(0, 0), c(1000, 0), c(1000, 1000), c(0, 1000), c(0, 0)
))), crs = 32633))

# Generate two covariate rasters aligned with the boundary
grid_template <- rast(nrows = 50, ncols = 50, xmin = 0, xmax = 1000, ymin = 0, ymax = 1000, crs = "EPSG:32633")

covariates <- list(
  fertility = setValues(grid_template, runif(ncell(grid_template), min = 0.2, max = 0.9)),
  moisture = setValues(grid_template, runif(ncell(grid_template), min = 0.1, max = 0.8))
)

field_data <- list(
  boundary = boundary,
  covariates = covariates,
  metadata = list(field_id = "Synthetic-Field", season = "2025")
)

existing_samples <- data.frame(
  x = c(100, 350, 700, 880, 150),
  y = c(120, 780, 640, 210, 450)
)

Initialize the tool 

tool <- create_ml_sampling_tool(
  config = list(
    log_level = "INFO"
  ),
  interactive = FALSE
)

Run BDL optimization with Uncertainty Quantification 

bdl_result <- tool$run_bdl(
  field_data = field_data,
  existing_samples = existing_samples,
  n_new_samples = 25,
  uncertainty_type = "total",
  mc_iterations = 50
)

# Inspect uncertainty metrics
print(bdl_result$uncertainties$uncertainty_summary)

Run Random Forest Optimization with Feature Importance 

rf_result <- tool$run_rf_optimization(
  field_data = field_data,
  existing_samples = existing_samples,
  n_new_samples = 25,
  feature_importance_method = "permutation"
)

# Check feature importance
print(rf_result$feature_importance)

Compare algorithms 

comparison <- tool$compare_models(
  field_data = field_data,
  existing_samples = existing_samples,
  n_new_samples = 25,
  algorithms = c("BDL", "RF", "udl"),
  n_iterations = 3
)

# Generate comparison report
tool$generate_ml_report(
  result = comparison,
  report_type = "comprehensive",
  output_dir = tempdir()
)

Next steps