## ----setup, include=FALSE-----------------------------------------------------
knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE)

## -----------------------------------------------------------------------------
library(spatialcvR)

# Example predictions vs observations
observed <- c(10, 15, 20, 25, 30, 35, 40, 45, 50, 55)
predicted <- c(11, 14, 21, 24, 31, 34, 41, 44, 51, 54)

# Calculate metrics
metrics <- spatial_metrics(observed, predicted)

print(metrics)

## -----------------------------------------------------------------------------
# Perfect prediction
observed_perfect <- c(10, 20, 30, 40, 50)
predicted_perfect <- c(10, 20, 30, 40, 50)
metrics_perfect <- spatial_metrics(observed_perfect, predicted_perfect)
print(metrics_perfect)

# Poor prediction
observed_poor <- c(10, 20, 30, 40, 50)
predicted_poor <- c(50, 40, 30, 20, 10)
metrics_poor <- spatial_metrics(observed_poor, predicted_poor)
print(metrics_poor)

## -----------------------------------------------------------------------------
# Load sample data
data(sample_spatial_data)

# Create spatial folds
folds_block <- spatial_folds(
  data = sample_spatial_data,
  x = "longitude",
  y = "latitude",
  k = 5,
  method = "block",
  seed = 123
)

# Create random folds for comparison
folds_random <- spatial_folds(
  data = sample_spatial_data,
  x = "longitude",
  y = "latitude",
  k = 5,
  method = "random",
  seed = 123
)

## -----------------------------------------------------------------------------
# Simulate model training and evaluation for each fold
evaluate_fold <- function(fold, data, target_var) {
  train_idx <- fold$train
  test_idx <- fold$test
  
  # Simple linear model for demonstration
  train_data <- data[train_idx, ]
  test_data <- data[test_idx, ]
  
  # Train model (using variable1 and variable2 as predictors)
  model <- lm(target ~ variable1 + variable2, data = train_data)
  
  # Make predictions
  predictions <- predict(model, newdata = test_data)
  observed <- test_data[[target_var]]
  
  # Calculate metrics
  metrics <- spatial_metrics(observed, predictions, na.rm = TRUE)
  
  return(metrics)
}

# Evaluate spatial block CV
block_results <- lapply(folds_block$folds, function(fold) {
  evaluate_fold(fold, sample_spatial_data, "target")
})

# Evaluate random CV
random_results <- lapply(folds_random$folds, function(fold) {
  evaluate_fold(fold, sample_spatial_data, "target")
})

## -----------------------------------------------------------------------------
# Aggregate metrics across folds
aggregate_metrics <- function(results_list) {
  rmse_values <- sapply(results_list, function(x) x$RMSE)
  mae_values <- sapply(results_list, function(x) x$MAE)
  r2_values <- sapply(results_list, function(x) x$R2)
  
  list(
    RMSE_mean = mean(rmse_values),
    RMSE_sd = sd(rmse_values),
    MAE_mean = mean(mae_values),
    MAE_sd = sd(mae_values),
    R2_mean = mean(r2_values),
    R2_sd = sd(r2_values)
  )
}

block_summary <- aggregate_metrics(block_results)
random_summary <- aggregate_metrics(random_results)

cat("Spatial Block CV Summary:\n")
print(block_summary)

cat("\nRandom CV Summary:\n")
print(random_summary)

## -----------------------------------------------------------------------------
# Create summary metrics for comparison
block_summary_metrics <- list(
  RMSE = block_summary$RMSE_mean,
  MAE = block_summary$MAE_mean,
  R2 = block_summary$R2_mean
)

random_summary_metrics <- list(
  RMSE = random_summary$RMSE_mean,
  MAE = random_summary$MAE_mean,
  R2 = random_summary$R2_mean
)

# Compare methods
comparison <- compare_cv(
  results_list = list(
    spatial_block = block_summary_metrics,
    random = random_summary_metrics
  )
)

print(comparison)

## ----eval=FALSE---------------------------------------------------------------
# # Get predictions from one fold for residual analysis
# fold_idx <- 1
# train_idx <- folds_block$folds[[fold_idx]]$train
# test_idx <- folds_block$folds[[fold_idx]]$test
# 
# train_data <- sample_spatial_data[train_idx, ]
# test_data <- sample_spatial_data[test_idx, ]
# 
# # Train model
# model <- lm(target ~ variable1 + variable2, data = train_data)
# 
# # Make predictions
# predictions <- predict(model, newdata = test_data)
# observed <- test_data$target
# 
# # Remove NA values
# valid_idx <- !is.na(predictions) & !is.na(observed)
# predictions <- predictions[valid_idx]
# observed <- observed[valid_idx]
# coords <- cbind(
#   x = test_data$longitude[valid_idx],
#   y = test_data$latitude[valid_idx]
# )
# 
# # Calculate spatial residuals
# residuals <- spatial_residuals(observed, predictions, coords)
# 
# print(residuals)

## ----eval=FALSE---------------------------------------------------------------
# # Spatial scatter plot of residuals
# plot_spatial_residuals(residuals, type = "scatter",
#                       main = "Spatial Residuals - Scatter Plot")
# 
# # Histogram of residuals
# # (Chunk disabled due to dependency on previous disabled chunk)
# # plot_spatial_residuals(residuals, type = "histogram",
# #                       main = "Spatial Residuals - Histogram")
# 
# # Q-Q plot for normality
# # (Chunk disabled due to dependency on previous disabled chunk)
# # plot_spatial_residuals(residuals, type = "qq",
# #                       main = "Spatial Residuals - Q-Q Plot")

## ----eval=FALSE---------------------------------------------------------------
# # Simulate agricultural data
# set.seed(456)
# n <- 100
# agri_data <- data.frame(
#   longitude = runif(n, 0, 1000),
#   latitude = runif(n, 0, 1000),
#   soil_quality = rnorm(n, 50, 10),
#   rainfall = rnorm(n, 800, 100),
#   temperature = rnorm(n, 20, 3),
#   yield = 50 + 0.3*soil_quality + 0.02*rainfall + 0.5*temperature +
#            rnorm(n, 0, 5)
# )
# 
# # Add spatial structure to yield
# agri_data$yield <- agri_data$yield +
#   0.01*agri_data$longitude + 0.02*agri_data$latitude
# 
# # Create spatial folds
# agri_folds <- spatial_folds(
#   data = agri_data,
#   x = "longitude",
#   y = "latitude",
#   k = 5,
#   method = "block",
#   seed = 456
# )
# 
# # Check for spatial leakage
# agri_leakage <- detect_spatial_leakage(
#   data = agri_data,
#   folds = agri_folds,
#   x = "longitude",
#   y = "latitude"
# )
# 
# print(agri_leakage)
# 
# # Evaluate model
# agri_results <- lapply(agri_folds$folds, function(fold) {
#   evaluate_fold(fold, agri_data, "yield")
# })
# 
# agri_summary <- aggregate_metrics(agri_results)
# print(agri_summary)

## ----eval=FALSE---------------------------------------------------------------
# # Compare multiple spatial CV methods
# methods <- c("block", "cluster", "random")
# method_results <- list()
# 
# for (method in methods) {
#   folds <- spatial_folds(
#     data = sample_spatial_data,
#     x = "longitude",
#     y = "latitude",
#     k = 5,
#     method = method,
#     seed = 123
#   )
# 
#   results <- lapply(folds$folds, function(fold) {
#     evaluate_fold(fold, sample_spatial_data, "target")
#   })
# 
#   summary <- aggregate_metrics(results)
#   method_results[[method]] <- list(
#     RMSE = summary$RMSE_mean,
#     MAE = summary$MAE_mean,
#     R2 = summary$R2_mean
#   )
# }
# 
# # Compare all methods
# multi_comparison <- compare_cv(method_results)
# print(multi_comparison)

## ----eval=FALSE---------------------------------------------------------------
# # Test different levels of spatial separation
# k_values <- c(3, 5, 7, 10)
# separation_results <- data.frame(
#   k = integer(),
#   mean_distance = numeric(),
#   RMSE = numeric()
#   # (Chunks disabled for CRAN build - they work interactively but not during build)
#   # For demonstration, please see the other vignettes or run the examples interactively
# )

## -----------------------------------------------------------------------------
# Establish baseline
random_baseline <- spatial_folds(sample_spatial_data, "longitude", "latitude",
                                k = 5, method = "random", seed = 123)

# Compare with spatial methods
spatial_methods <- spatial_folds(sample_spatial_data, "longitude", "latitude",
                                k = 5, method = "block", seed = 123)

## -----------------------------------------------------------------------------
# Always check leakage before trusting results
leakage_check <- detect_spatial_leakage(sample_spatial_data, spatial_methods,
                                       "longitude", "latitude")
if (leakage_check$risk_level == "high") {
  warning("High spatial leakage detected - results may be unreliable")
}

## -----------------------------------------------------------------------------
# Don't just look at aggregate metrics
# Examine spatial patterns in errors
# This can reveal missing spatial predictors

## -----------------------------------------------------------------------------
# Keep track of:
# - CV method used
# - Parameters (k, block size, buffer radius, etc.)
# - Spatial leakage assessment
# - Performance metrics
# - Residual analysis results