2D contour heatmaps for illustrating factor relationships in machine learning

plot_class_distrib2D.R

require(mlbench)
require(rgl)
require(MASS)
require(plot3D)

##' diverging colour palette function with set midpoint
##'
##' returns a palette function that maps values to colours, with
##' a midpoint (defaulting to 0) corresponding to the central colour
##' @title diverging_palette
##' @export
##' @param d data giving the range of the palette
##' @param centered logical, whether to use both sides from the midpoint symmetrically
##' @param midpoint numeric value corresponding to the central colour
##' @param colors vector of colors, length must be odd
##' @return a function
##' @author baptiste Auguie
##' @family low_level
##' @examples
##' grid.raster(diverging_palette(1:10, TRUE, mid=2, col=c("blue", "white", "red"))(1:10))
diverging_palette <- function(d = NULL, centered = FALSE, midpoint = 0,
                              colors = RColorBrewer::brewer.pal(7,"PRGn")){

  half <- length(colors)/2

  if(!length(colors)%%2)
    stop("requires odd number of colors")
  if( !centered && !(midpoint <=  max(d, na.rm=TRUE) && midpoint >= min(d, na.rm=TRUE)))
    warning("Midpoint is outside the data range!")
  
  values <-  if(!centered) {
    low <- seq(min(d, na.rm=TRUE), midpoint, length=half)
    high <- seq(midpoint, max(d, na.rm=TRUE), length=half)
    c(low[-length(low)], midpoint, high[-1])
  } else {
    mabs <- max(abs(d - midpoint), na.rm=TRUE)
    seq(midpoint-mabs, midpoint + mabs, length=length(colors))
  }
  
  scales::gradient_n_pal(colors, values = values)
  
}

data(PimaIndiansDiabetes)

data_p <- subset(PimaIndiansDiabetes,diabetes=='pos')
data_n <- subset(PimaIndiansDiabetes,diabetes=='neg')

kde_p <- kde2d(data_p$glucose,data_p$mass,n=500)
kde_n <- kde2d(data_n$glucose,data_n$mass,n=500)

z <- kde_p$z - kde_n$z
colors <- diverging_palette( d=z, 
                              colors=RColorBrewer::brewer.pal(11,"Spectral") )(seq(min(z),max(z),(max(z)-min(z))/10))

image2D(x = kde_p$x, y = kde_p$y, z, 
                              col = colors,
                              xlab="glucose",ylab="mass",
                              shade = 0.2, rasterImage = TRUE, 
                              contour = list(col = "white", labcex = 0.8, lwd = 1, alpha = 0.5))