R/alfa.rf.R

In CompositionalML: Machine Learning with Compositional Data

alfa.rf <- function(xnew, y, x, a = seq(-1, 1, by = 0.1), size = c(1, 2, 3),
                    depth = c(0, 1), splits = 2:5, R = 500) {

  if ( min(x) == 0 )  a <- a[a > 0]

  config <- as.matrix( expand.grid(size = size, depth = depth, splits = splits, R = R) )

  mod <- list()
  for ( k in 1:length(a) ) {
    z <- Compositional::alfa(x, a[k])$aff
    z <- as.data.frame(z)
    xnew <- as.matrix(xnew)
    if ( dim(xnew)[1] == 1 )  xnew <- t(xnew)
    znew <- Compositional::alfa(xnew, a[k])$aff
    znew <- as.data.frame(znew)
    mod[[ k ]] <- .rfmodel(znew, y, z, config = config)
  }
  names(mod) <- paste("alpha=", a, sep = "")
  mod
}



.rfmodel <- function(xnew, y, x, config = config) {

  p <- dim(config)[1]
  mod <- list()
  est <- matrix(nrow = dim(xnew)[1], ncol = p)
  xnew <- as.data.frame(xnew)
  colnames(xnew) <- colnames(x)

  for ( j in 1:p ) {
    mod[[ j ]] <- ranger::ranger(y ~., data = x, importance = "impurity_corrected",
                          min.node.size = config[j, 1], splitrule = "extratrees", max.depth = config[j, 2],
                          num.random.splits = config[j, 3], num.trees = config[j, 4])
    est[, j] <- predict(mod[[ j ]], data = xnew)$predictions
  }  ##  end  for ( j in dim(config)[1] ) {

  list(mod = mod, config = config, est = est)
}