R/lfda_visualization.R

In lfda: Local Fisher Discriminant Analysis

#' Assigning Colors to A Vector
#'
#' This function assigns a color to each distinct value in the given vector.
#'
#' @param vec The vector where each distinct value will be assigned a color.
#'
#' @return The colors for each element in the given vector
#'
#' @import grDevices
#'
#' @export
#'
Cols <- function(vec) {
  cols <- grDevices::rainbow(length(unique(vec)))
  return(cols[as.numeric(as.factor(vec))])
}
#' 3D Visualization for LFDA/KLFDA Result
#'
#' This function plot 3 dimensions of the lfda/klfda result.
#'
#' @param x The lfda/klfda result.
#' @param labels A list of class labels used for lfda/klfda training.
#' @param cleanText A boolean value to specify whether to make the labels in the plot cleaner (default: FALSE)
#' @param ... Additional arguments
#'
#' @method plot lfda
#' @export
#'
#' @seealso See \code{\link{lfda}} and \code{\link{klfda}} for the metric learning method used for this visualization.
#'
plot.lfda <- function(x, labels, cleanText = FALSE, ...) {
  if (is.character(labels) | is.numeric(labels) | is.integer(labels)) {
    labels <- as.factor(labels)
  }
  if (!is.factor(labels)) {
    stop("labels need to be an object convertable to a factor.")
  }

  if (length(labels) != dim(x$Z)[1]) {
    stop("length of labels differs from the number of rows in x$Z.")
  }

  if (!is.logical(cleanText)) {
    stop("cleanText needs to be TRUE(T) or FALSE(F).")
  }

  if (dim(x$Z)[2] > 3) {
    x$Z <- x$Z[, 1:3]
    warning("dimensionality of the transformed data is larger than 3. Only the first 3 columns of the data will be used. ")
  } else if (dim(x$Z)[2] < 3) {
    stop("dimensionality of the transformed data must be larger than 3.")
  }

  transformedData <- as.data.frame(cbind(labels, x$Z))
  colnames(transformedData)[1] <- "Class"

  if (cleanText) {
    ## Show The Text of Each Style Only Once ##
    newData <- suppressWarnings( # known warnings
      plyr::ddply(transformedData, plyr::.(Class), function(y) {
        y[-1, "Class"] <- ""
        return(y)
      })
    )
    df <- sapply(newData[, "Class"], as.character)
    df[is.na(df)] <- "."
    newData[, "Class"] <- df
  } else {
    newData <- transformedData
  }

  # options(rgl.useNULL=TRUE) # deal with rgl in Travis

  ## Plot 3D Visualization of LFDA Result ##
  rgl::text3d(newData[, 2], newData[, 3], newData[, 4],
    col = Cols(transformedData$Class),
    main = "3D Visualization of Metric Transformed Data", size = 4,
    texts = newData$Class, font = 1, cex = 1.2, ...
  )
  rgl::axes3d()
}