Marginal Plots¶

A marginal plot is a scatterplot (sometimes a density plot or other bivariate plot) that has histograms, boxplots, or other distribution visualization layers in the margins of the x- and y-axes.

It allows studying the relationship between 2 numeric variables.

You can use the ggmarginal() function to add marginal layers to a plot.

In [1]:

%useLatestDescriptors
%use dataframe
%use lets-plot

In [2]:

@file:DependsOn("org.apache.commons:commons-math3:3.6.1")

In [3]:

import org.apache.commons.math3.distribution.MultivariateNormalDistribution

In [4]:

val cov0 : Array<DoubleArray> = arrayOf(doubleArrayOf(1.0, -.8),
                                        doubleArrayOf(-.8, 1.0))

val cov1 : Array<DoubleArray> = arrayOf(doubleArrayOf(10.0, .1),
                                        doubleArrayOf(.1, .1))

In [5]:

val n = 200

val means0 : DoubleArray = doubleArrayOf(-2.0, 0.0)
val means1 : DoubleArray = doubleArrayOf(0.0, 1.0)

val xy0 = MultivariateNormalDistribution(means0, cov0).sample(n)
val xy1 = MultivariateNormalDistribution(means1, cov1).sample(n)

In [6]:

val dat = mapOf(
    "x" to (xy0.map { it[0] } + xy1.map { it[0] }).toList(),
    "y" to (xy0.map { it[1] } + xy1.map { it[1] }).toList(),
    "c" to List(n){"A"} + List(n){"B"},
)

In [7]:

// The default plot theme and palette.
LetsPlot.theme = themeLight()
val paletteColors = listOf("#394449", "#F7C443")
val palette = scaleColorManual(values = paletteColors) + scaleFillManual(values = paletteColors)

In [8]:

val p = letsPlot(dat.toMap()){x = "x"; y = "y"; color = "c"; fill = "c"} + geomPoint() + palette
p

Out[8]:

In [9]:

// Add marginal layers to any side on the plot: 
// - [l]eft 
// - [r]ight 
// - [t]op 
// - [b]ottom 

p + ggmarginal("tr", layer = geomHistogram(color = "white")) +
    ggmarginal("lb", layer = geomDensity(size = 0))

Out[9]:

In [10]:

// Set constant colors to disable the grouping.

val pHist = p +
            ggmarginal("tr", layer=geomHistogram(fill = "gray", color = "white")) +
            ggmarginal("tr", layer=geomDensity(fill = "rgba(0,0,0,0)", color = "red"))
pHist

Out[10]:

In [11]:

// Add marginal boxplots.
// Make margins smaller using the parameter `size`.

pHist + ggmarginal("lb", size=0.03, layer=geomBoxplot(fill = "white", color = "black"))

Out[11]:

In [12]:

// Combine marginal violin and boxplot.
// Make margins much larger.

p + ggmarginal("rb", size=0.4, layer=geomViolin(trim = false, color = "black")) +
    ggmarginal("rb", layer=geomBoxplot(mapping = {group = "c"}, fill = "white", color = "white", 
                                        alpha = .25, outlierColor = "red", width = .2))

Out[12]:

In [13]:

// Show marginal densities on a 2D-density plot.

letsPlot(dat) {x = "x"; y = "y"} + geomDensity2DFilled {fill = "..level.."} + coordCartesian() +
 ggmarginal("tr", layer = geomArea(stat = Stat.density()))

Out[13]:

In [14]:

var mpg = DataFrame.readCSV("https://raw.githubusercontent.com/JetBrains/lets-plot-kotlin/master/docs/examples/data/mpg.csv")
mpg.head(3)

Out[14]:

DataFrame: rowsCount = 3, columnsCount = 12

untitled	manufacturer	model	displ	year	cyl	trans	drv	cty	hwy	fl	class
1	audi	a4	1.800000	1999	4	auto(l5)	f	18	29	p	compact
2	audi	a4	1.800000	1999	4	manual(m5)	f	21	29	p	compact
3	audi	a4	2.000000	2008	4	manual(m6)	f	20	31	p	compact

In [15]:

// A marginal plot with a significant number of groups on margins.

val plotSettings = ggsize(900, 700) +
                   theme(plotBackground = elementRect(fill = "#eaeaea"),
                         legendBackground = elementRect(fill = "#eaeaea"))

(letsPlot(mpg.toMap()) {x = "cty"; y = "hwy"; fill = "manufacturer"}
    + geomPoint( size = 7, shape = 21, alpha = .5, color = "black", position = positionJitter(seed = 42))
    + ggmarginal("tr", size = 0.3, 
                 layer = geomBoxplot(width=0.5) {
                     fill = asDiscrete("manufacturer", orderBy = "..lower..")
                 })
    + plotSettings)

Out[15]: