R/RunPredictLeaf.R

In rerf: Randomer Forest

#' Calculate similarity using a single tree.
#'
#' This is the base function called by ComputeSimilarity.
#'
#' @param X an n sample by d feature matrix (preferable) or data frame which was used to train the provided forest.
#' @param tree a tree from a forest returned by RerF.
#'
#' @return similarity based on one tree
#'


RunPredictLeaf <-
  function(X, tree) {
    tm <- 0L
    currentNode <- 0L
    curr_ind <- 0L
    classProb <- double(length(tree$ClassProb[1L, ]))
    num_classes <- ncol(tree$ClassProb)
    n <- nrow(X)

    # do we need to rotate the data?
    if (!is.null(tree$rotmat)) {
      if (is.null(tree$rotdims)) {
        X[] <- X %*% tree$rotmat
      } else {
        X[, tree$rotdims] <- X[, tree$rotdims] %*% tree$rotmat
      }
    }

    leafIdx <- integer(n)
    Xnode <- double(n)
    numNodes <- length(tree$treeMap)
    Assigned2Node <- vector("list", numNodes)
    Assigned2Node[[1L]] <- 1L:n
    for (m in 1L:numNodes) {
      nodeSize <- length(Assigned2Node[[m]])
      if (nodeSize > 0L) {
        if ((tm <- tree$treeMap[m]) > 0L) {
          indexHigh <- tree$matAindex[tm + 1L]
          indexLow <- tree$matAindex[tm] + 1L
          s <- (indexHigh - indexLow + 1L) / 2L
          Xnode[1L:nodeSize] <- X[Assigned2Node[[m]], tree$matAstore[indexLow:indexHigh][(1L:s) * 2L - 1L], drop = FALSE] %*%
            tree$matAstore[indexLow:indexHigh][(1L:s) * 2L]
          moveLeft <- Xnode[1L:nodeSize] <= tree$CutPoint[tm]
          Assigned2Node[[tm * 2L]] <- Assigned2Node[[m]][moveLeft]
          Assigned2Node[[tm * 2L + 1L]] <- Assigned2Node[[m]][!moveLeft]
        } else {
          leafIdx[Assigned2Node[[m]]] <- tm * -1L
        }
      }
      Assigned2Node[[m]] <- NA
    }
    return(leafIdx)
  }