R/learn-params-awnb.R

In bmihaljevic/bnclassify: Learning Discrete Bayesian Network Classifiers from Data

awnb <- function(class, dataset, trees = NULL, bootstrap_size = NULL) {
  if (is.null(trees)) trees <- 10
  if (is.null(bootstrap_size)) bootstrap_size <- 0.5
# For each tree, get a bootstrap subsample 
  subsamples <- replicate(trees, bootstrap_ss(dataset, bootstrap_size), 
                          simplify = FALSE)
# From each sample, learn tree
  Wtrees <- lapply(subsamples, learn_unprunned_tree, class)
# For each tree get minimum testing depth 
  depths <- lapply(Wtrees, identify_min_testing_depths)
  depths <- unlist(depths, use.names = TRUE)
  if (length(depths) == 0) stop("Only empty trees have been learned.")
  features <- get_features(class, dataset)
  unused_features <- features[!(features %in% names(depths))]
  depths[unused_features] <- Inf
# Compute weights and average across the features
  tapply(depths, names(depths), 
         function(x) sum(x ^ -0.5), simplify = TRUE) / trees
}