R/ForestRegression.R

In mnwright/simpleRF: Random Forests with emphasis on simplicity

##' @title Regression forest class
##' @description Subclass for regression forest. 
##' Contains all fields and methods used special for regression forests.
ForestRegression <- setRefClass("ForestRegression", 
  contains = "Forest", 
  fields = list(), 
  methods = list(
    
    grow = function(num_threads) {      
      treetype <<- "Regression"
      
      ## Create trees
      trees <<- replicate(num_trees, TreeRegression$new())

      ## Call parent method
      callSuper(num_threads)
    },
    
    predict = function(newdata) {
      callSuper(newdata)
    },
    
    aggregatePredictions = function(predictions) {
      ## For all samples take mean over all trees
      return(rowMeans(predictions))
    }, 
    
    predictionError = function() {
      ## For each tree loop over OOB samples and get predictions
      tree_predictions <- sapply(trees, function(x) {
        oob_samples <- x$oob_sampleIDs
        result <- rep(NA, data$nrow)
        result[oob_samples] <- x$predictOOB()
        return(result)
      })
      
      ## Compute prediction for each sample
      predicted_response <- rowMeans(tree_predictions, na.rm = TRUE)
      
      ## Compare predictions with true data
      return(sum((predicted_response - data$column(1))^2, na.rm = TRUE) / data$nrow)
    })
)