# Install required libraries
install.packages("ranger")
install.packages("iml")

# Load the libraries
library(ranger)
library(iml)

# Define the target
targetTag <- "Petal.Width"

# Training dataset
irisNUM <- iris[2:nrow(iris), 1:(ncol(iris) - 1)]

# Record to explain using Shapley values
irisNUM_explain <- iris[1, 1:3]

# Adjust the weights based on recent observations, to give it more weight
weight <- c(1:nrow(irisNUM))
weight <- round(exp(weight / (length(weight) / 3)), digits = 1) 

# Define the formula for the model
targetForm <- paste(targetTag, " ~ .", sep = "")

# Train the model
set.seed(123)
model <- ranger(as.formula(targetForm), data = irisNUM, write.forest = TRUE, 
                min.node.size = 25, case.weights = weight, seed = 123)

# Create custom function for the predictor
predFunction <- function(object, newdata) predict(object, data = newdata)$predictions

# Create the predictor 
predictor <- Predictor$new(model = model, 
                           data = irisNUM[ , 1:3], 
                           y = irisNUM[ , targetTag], 
                           predict.function = predFunction)

# Calculater the shapley values
shapley <- Shapley$new(predictor, 
                       x.interest = irisNUM_explain)

# Sum of Shapley values
SHAP_sum <- sum(shapley$results$phi)

# The difference the sum of Shapley values should be equal to
diff <- shapley$y.hat.interest - shapley$y.hat.average

# Final check, must be TRUE
SHAP_sum == diff