ranger

ranger

Fast random forests.

Basic Usage

library(ranger)

# Classification
model <- ranger(
  formula = target ~ .,
  data = train,
  num.trees = 500,
  importance = "impurity"
)

# Regression
model <- ranger(
  formula = value ~ .,
  data = train,
  num.trees = 500
)

# Predict
pred <- predict(model, test)
pred$predictions

Parameters

model <- ranger(
  formula = target ~ .,
  data = train,
  
  # Trees
  num.trees = 500,                   # Number of trees
  mtry = NULL,                       # Variables per split (default: sqrt(p))
  min.node.size = 1,                 # Min node size (1 class, 5 reg)
  max.depth = NULL,                  # Max depth (NULL = unlimited)
  
  # Sampling
  sample.fraction = 1,               # Sample fraction
  replace = TRUE,                    # Sample with replacement
  case.weights = NULL,               # Case weights
  
  # Importance
  importance = "none",               # none, impurity, impurity_corrected, permutation
  
  # Probability
  probability = FALSE,               # Probability forest
  
  # Other
  num.threads = NULL,                # Threads (NULL = all)
  seed = NULL,                       # Random seed
  verbose = TRUE,                    # Verbose output
  write.forest = TRUE                # Save forest
)

Probability Prediction

# Train probability forest
model <- ranger(
  formula = target ~ .,
  data = train,
  probability = TRUE
)

# Predict probabilities
pred <- predict(model, test)
pred$predictions  # Matrix of probabilities

Feature Importance

# Impurity importance
model <- ranger(target ~ ., data = train, importance = "impurity")
importance(model)

# Permutation importance
model <- ranger(target ~ ., data = train, importance = "permutation")
importance(model)

# Plot
barplot(sort(importance(model), decreasing = TRUE)[1:20])

Survival Analysis

library(survival)

# Survival forest
model <- ranger(
  formula = Surv(time, status) ~ .,
  data = train,
  num.trees = 500
)

# Predict survival
pred <- predict(model, test)
pred$survival  # Survival probabilities
pred$unique.death.times  # Time points