KDE scatterplot

kde_scatterplot.R

library(data.table)
library(ggplot2)
library(ks)

# This assumes that you have a data.table, stored in `dt.data`, with columns `x` and `y`.

get.densities <- function (x, y) {
    
    # You can also define your own bandwidth matrix or generate one using Hpi().
    
	bw.x <- abs(do.call('-', as.list(range(x)))) / 10
	bw.y <- abs(do.call('-', as.list(range(y)))) / 10
	H <- diag(c(bw.x, bw.y), nrow = 2, ncol = 2)
    
	kde.res <- kde(cbind(x, y), eval.points = cbind(x, y), H = H)
	kde.res$estimate / max(kde.res$estimate)
    
}

dt.data[, density := get.densities(x, y)]

# Ensure that higher densities are plotted last (i.e., on top of everything else).

setorder(dt.data, density)

# Drop duplicate x and y values -- plotting them is pointless.

dt.data <- unique(dt.data, by = c('x', 'y'))

# Plot away. Remember to add a colour scale that you like!

ggplot(dt.data, aes(x = x, y = y)) + geom_point(aes(colour = density)) + scale_colour_gradientn(colours = c(grey(0.85), 'black'), breaks = pretty)