R/d_KPCA.R

In egenn/rtemis: Machine Learning and Visualization

# d_KPCA.R
# ::rtemis::
# 2016 E.D. Gennatas www.lambdamd.org

#' Kernel Principal Component Analysis
#'
#' Perform kernel PCA decomposition using `kernlab::kpca`
#'
#' Project scaled variables to KPCA components.
#' Input must be n by p,
#' where n represents number of cases,
#' and p represents number of features.
#' KPCA will be applied to the transpose of the n x p matrix.
#'
#' @inheritParams d_SVD
#' @param x Input data
#' @param x.test Optional test set. Will be projected on to KPCA basis
#' @param k Integer vector of length 1 or greater. N of components to return
#'   If set to 0, `th` determines eigenvalue below which PCs are ignored
#' @param th Threshold for eigenvalue below which PCs are ignored if `k` is set to 0
#' @param kernel Character: Type of kernel to use. See `kernlab::kpca`
#' @param kpar List of hyperparameters: See `kernlab::kpca("kpar")`
#' @param center Logical: If TRUE, center data prior to decomposition. Default = TRUE
#' @param scale Logical: If TRUE, scale data prior to decomposition. Default = TRUE
#' @param ... Additional parameters to be passed to `fastKPCA::fastKPCA`
#' @return `rtDecom` object
#' @author E.D. Gennatas
#' @family Decomposition
#' @export

d_KPCA <- function(x,
                   x.test = NULL,
                   k = 2,
                   th = 0.0001,
                   kernel = "rbfdot",
                   kpar = NULL,
                   center = TRUE,
                   scale = TRUE,
                   verbose = TRUE, ...) {

  # Intro ----
  start.time <- intro(verbose = verbose)
  decom.name <- "KPCA"

  # Dependencies ----
  dependency_check("kernlab")

  # Arguments ----
  # Kernel parameters
  if (is.null(kpar)) {
    if (kernel == "rbfdot") {
      kpar <- list(sigma = 0.1)
    }
  }

  # Data ----
  n <- NROW(x)
  p <- NCOL(x)
  if (verbose) {
    msg2("||| Input has dimensions ", n, " rows by ", p, " columns,", sep = "")
    msg2("    interpreted as", n, "cases with", p, "features.")
  }
  if (is.null(colnames(x))) colnames(x) <- paste0("Feature_", seq_len(NCOL(x)))
  xnames <- colnames(x)
  if (!is.null(x.test)) colnames(x.test) <- xnames
  x <- as.matrix(x)

  # scale ----
  if (scale || center) {
    x <- scale(x, scale = scale, center = center)
    .center <- attr(x, "scaled:center")
    .scale <- attr(x, "scaled:scale")
  } else {
    .center <- .scale <- FALSE
  }

  # KPCA ----
  if (verbose) msg2("Running Kernel Principal Components Analysis...")
  decom <- kernlab::kpca(x, features = k, th = th,
                         kernel = kernel, kpar = kpar, ...)
  vectors <- decom@pcv

  # Projections ----
  projections.test <- NULL
  projections.train <- kernlab::predict(decom, x)
  if (!is.null(x.test)) {
    if (scale || center) {
      # x.test <- (x.test + .center) %*% diag(.scale) # faster for small matrices only
      x.test <- t(t(x.test + .center) * .scale)
    }
    projections.test <- kernlab::predict(decom, x.test)
  }

  # Outro ----
  extra <- list(vectors = vectors)
  rt <- rtDecom$new(decom.name = decom.name,
                    decom = decom,
                    xnames = xnames,
                    projections.train = projections.train,
                    projections.test = projections.test,
                    parameters = list(k = k,
                                      th = th,
                                      kernel = kernel,
                                      kpar = kpar,
                                      scale = scale,
                                      center = center),
                    center = .center,
                    scale = .scale,
                    extra = extra)
  outro(start.time, verbose = verbose)
  rt

} # rtemis::d_KPCA