15 Distributions
Histograms, density plots, boxplots, violins, and ridgelines for exploring the shape and spread of data.
(ns plotje-book.distributions
(:require
;; Rdatasets -- standard datasets
[scicloj.metamorph.ml.rdatasets :as rdatasets]
;; Kindly -- notebook rendering protocol
[scicloj.kindly.v4.kind :as kind]
;; Plotje -- composable plotting
[scicloj.plotje.api :as pj]))Histogram
Distribution of sepal length across all species.
(-> (rdatasets/datasets-iris)
(pj/lay-histogram :sepal-length))Horizontal Histogram
Naming the column on :y draws the histogram on its side, with the counts along :x. It is the plot the column on :x draws with the coordinates flipped.
(-> (rdatasets/datasets-iris)
(pj/lay-histogram {:y :sepal-length}))Colored Histogram
Split by species β each group gets its own color.
(-> (rdatasets/datasets-iris)
(pj/lay-histogram :sepal-length {:color :species}))Petal Width Histogram
Petal width has a bimodal distribution.
(-> (rdatasets/datasets-iris)
(pj/lay-histogram :petal-width))Histogram with Custom Title
(-> (rdatasets/reshape2-tips)
(pj/lay-histogram :total-bill)
(pj/options {:title "Distribution of Total Bill"
:x-label "Amount ($)"}))Density-Normalized Histogram
Pass {:normalize :density} so the y-axis shows probability density instead of raw counts. This makes the histogram directly comparable with a density curve drawn on the same panel.
(-> (rdatasets/datasets-iris)
(pj/lay-histogram :sepal-length {:normalize :density :alpha 0.5})
pj/lay-density)Log-Scale Histogram
When bin counts span orders of magnitude, the smallest bars disappear next to the largest on a linear y-axis. A log y-scale lets every bar register. The data below doubles the count of each successive bin (1, 2, 4, β¦, 512); on a log axis the doubling shows as a uniform staircase β each bar a fixed step above the previous, the same step every time.
The lower bound under log comes from the smallest positive bin count, not from the visual zero baseline β log scales have no zero. Empty bins emit no bar and do not pull the axis down.
(-> {:x (mapcat (fn [i] (repeat (long (Math/pow 2 i)) i)) (range 10))}
(pj/lay-histogram {:bins 10})
(pj/scale :y :log)
(pj/options {:title "Log Y on Histogram"}))Density Plot
A smooth curve estimating the probability density function. Less sensitive to bin width than histograms.
(-> (rdatasets/datasets-iris)
(pj/lay-density :sepal-length))Grouped Density
Per-species density curves with automatic color mapping.
(-> (rdatasets/datasets-iris)
(pj/lay-density :sepal-length {:color :species}))Each curve is estimated over the range of every speciesβ values, not only its own, so all three share one interval. A curve falls away to nothing at an end its speciesβ values stop short of. Setosa holds the smallest sepal length and virginica the largest, so the setosa curve is cut off at the left end of the interval and the virginica curve at the right end. Pass {:trim true} to estimate each species over its own values instead. Trimming shows where each groupβs data lies, at the cost of cutting each curve off at its extremes.
(-> (rdatasets/datasets-iris)
(pj/lay-density :sepal-length {:color :species :trim true}))Density with Custom Bandwidth
Bandwidth sets how heavily the curve is smoothed. Left alone it follows the rule Rβs density() uses, which for this column works out near 0.27. A narrower bandwidth reveals more detail β the curve below rises higher and breaks into bumps the default smooths away β while a wider one flattens it further.
(-> (rdatasets/datasets-iris)
(pj/lay-density :sepal-length {:bandwidth 0.1}))Density with Fill and Outline
By default a density fills the area under the curve in the mapped color. To make the curveβs shape read more crisply, add an outline with :stroke. The fill still comes from :color, and the outline traces only the top curve, not the baseline. :stroke-width sets the outline thickness.
(-> (rdatasets/datasets-iris)
(pj/lay-density :sepal-length {:color "lightblue" :stroke "black" :stroke-width 2}))One filled polygon plus one outline line, and the render carries both the light-blue fill and the black outline color.
Rug
A rug shows each observationβs value as a short tick mark along the axis. Layered with a density curve, it shows the smooth shape and the underlying observations together.
The curve and the rug start and stop together. A density estimate is defined everywhere, not only where the data is, so drawing one means choosing an interval to estimate it over, and Plotje uses the observed values. Both therefore stop a little short of the axis ends, which carry the same padding every numeric axis gets. This matches what ggplot2βs geom_density() draws for the same data.
(-> (rdatasets/datasets-iris)
(pj/lay-density :sepal-length)
pj/lay-rug)Strip Plot (Jitter)
When plotting a numeric column against a categorical column, points fall on the same place in each band. :jitter true spreads them with small random offsets along the categorical axis.
(-> (rdatasets/datasets-iris)
(pj/lay-point :species :sepal-width {:jitter true}))Pass a number to control the jitter amount in drawing units.
(-> (rdatasets/datasets-iris)
(pj/lay-point :species :sepal-width {:jitter 10 :alpha 0.5}))Boxplot
Median, quartiles, whiskers at 1.5 times the IQR (interquartile range), and outlier points.
(-> (rdatasets/datasets-iris)
(pj/lay-boxplot :species :sepal-width))Each whisker stays within the Tukey fence [Q1 - 1.5*IQR, Q3 + 1.5*IQR], and every outlier falls outside it.
Grouped Boxplot
Side-by-side boxplots colored by a grouping variable.
(-> (rdatasets/reshape2-tips)
(pj/lay-boxplot :day :total-bill {:color :smoker}))Each color group gets a distinct dodge offset, visible as side-by-side boxes within each day.
Horizontal Boxplot
Flipped coordinate for horizontal orientation.
(-> (rdatasets/datasets-iris)
(pj/lay-boxplot :species :sepal-width)
(pj/coord :flip))Violin Plot
A violin shows the full density shape per category β more informative than a boxplot for multimodal distributions.
(-> (rdatasets/reshape2-tips)
(pj/lay-violin :day :total-bill))A violin takes the opposite default to a density: each body is estimated over its own categoryβs values, so it ends where that dayβs bills end rather than tapering past them. Pass {:trim false} to let the tails fall away instead.
Grouped Violin
Color splits each category into side-by-side violins.
(-> (rdatasets/reshape2-tips)
(pj/lay-violin :day :total-bill {:color :smoker}))Each color group gets a distinct dodge offset, visible as side-by-side violins within each day.
Horizontal Violin
(-> (rdatasets/datasets-iris)
(pj/lay-violin :species :petal-length)
(pj/coord :flip))Ridgeline Plot
Overlapping density curves stacked vertically by category β good for comparing distribution shapes across many groups. The category is given first and drawn on the vertical axis, with the measurement along the horizontal axis β the reverse of ggridgesβ aes(x = value, y = group).
(-> (rdatasets/datasets-iris)
(pj/lay-ridgeline :species :sepal-length))Colored Ridgeline
Map color to the same categorical column for distinct curves.
(-> (rdatasets/datasets-iris)
(pj/lay-ridgeline :species :sepal-length {:color :species}))Comparing Multiple Columns
Name several columns in the mapping, {:x [...]}, to read them as a series on one panel, a color per column β here with pj/lay-histogram.
(pj/lay-histogram (rdatasets/datasets-iris)
{:x [:sepal-length :sepal-width :petal-length]})Use pj/facet over the columns to create one panel per column. This is useful for comparing the shape of different variables side by side.
(-> (rdatasets/datasets-iris)
(pj/facet {:x [:sepal-length :sepal-width :petal-length]})
(pj/lay-histogram))A :color column written beside the several columns of a series replaces the color per column. {:x [:sepal-length :sepal-width :petal-length] :color :species} colors the bars by species, and the legend lists the species and no longer the columns, so the three columns cannot be told apart. To see both, give each column its own panel and color by species within it:
(-> (rdatasets/datasets-iris)
(pj/facet {:x [:sepal-length :sepal-width :petal-length]})
(pj/lay-density {:color :species}))Several columns on :x suit the layer types that draw from one column, such as histograms, densities and rugs. A boxplot or a violin draws from two columns, so for those layer types the several columns go on :y, beside a category column on :x:
(pj/lay-boxplot (rdatasets/datasets-iris)
{:x :species
:y [:sepal-length :sepal-width :petal-length]})See Also
- Core Concepts β mappings and aesthetics referenced throughout
- Relationships β two-distribution comparisons via heatmap, contour, and SPLOM, and a distribution drawn beside a scatter