Plotnine - a comparison of 1D density graphs

Mon 09 March 2026

Plotnine and 1D density¶

Introduction¶

While surfing the web, I came across a graphic produced by ggplot2 in R, and I wondered if I could replicate it in plotnine and Python.

The ovetall graphic shows a row of tall thing graphics, each a different visualization of the density of a set of points on a line. I chose a bi-modal two-Gaussian dataset to visualize. To get my tall, thin plots, in most cases I rotate my first-pass graphic, which I create with the minimal code possible

Implementation¶

Imports¶

numpy handle numerical cxomputing
pandas handles dataframes
plotnine does all the plotting
warnings can be used to supress usere warning from plotnine

In [1]:

import warnings

import numpy as np
import pandas as pd
import plotnine as p9

Magics¶

watermark produces reproducability information

In [2]:

%load_ext watermark

Create dataset¶

We get 2000 points, 1000 each from two different Gaussians, and create a pandas DataFrame. The fill column holds the text name of our desired fill color (if fill is supported by the graphic style)

In [3]:

size = 1000
x = list(np.random.normal(-2, 1, size)) + list(np.random.normal(2, 1, size))

In [4]:

df = pd.DataFrame({"x": x, "fill": ["white"] * (2 * size)})

Graphics¶

Point plot¶

The first approach is to plot the data points as graphic points, with a very high transparency (low alpha). The overlapping points show the density of the dataset. We will label every plotnine plot with a descriptive subtitle. Every plot will have a black&white theme

In [5]:

plot = (
    p9.ggplot(data=df)
    + p9.geom_point(
        mapping=p9.aes(x="x", y=0),
        color="gray",
        alpha=0.01,
        size=5,
    )
    + p9.theme_bw()
    + p9.labs(subtitle="point")
)
plot

No description has been provided for this image

Flip axis to get a vertical graphic

In [6]:

plot2 = plot + p9.coord_flip()
plot2

Histogram¶

We set bins to 40, and set the outer boundary of each bar to be gray

In [7]:

plot3 = (
    p9.ggplot(data=df)
    + p9.geom_histogram(
        mapping=p9.aes(x="x"),
        color="gray",
        bins=40,
    )
    + p9.theme_bw()
    + p9.labs(subtitle="histogram")
)
plot3

Flip axis to get a vertical graphic

In [8]:

plot4 = plot3 + p9.coord_flip()
plot4

sina¶

This is a specialized graphic, similar to violin plots. It produces a density curve, filled with random jittered dots.

p9.scale_fill_identity() says treat the fill variable as a color name, not as some discrete value.

The x axis labels and tick marks aren't very meaningful, so they are removed

In [9]:

plot5 = (
    p9.ggplot(data=df)
    + p9.geom_sina(
        mapping=p9.aes(x=0, y=x, fill="fill"),
        maxwidth=0.1,
    )
    + p9.scale_fill_identity()
    + p9.theme_bw()
    + p9.labs(subtitle="sina")
)
plot5 + p9.theme(
    axis_title_x=p9.element_blank(),  # Removes the y-axis title
    axis_text_x=p9.element_blank(),  # Removes the y-axis tick labels
    axis_ticks_x=p9.element_blank(),  # Turns off y-axis ticks
)

Flip axis to get a horizontal graphic (not used subsequently)

In [10]:

plot6 = (
    plot5
    + p9.coord_flip()
    + p9.theme(
        axis_title_y=p9.element_blank(),  # Removes the y-axis title
        axis_text_y=p9.element_blank(),  # Removes the y-axis tick labels
        axis_ticks_y=p9.element_blank(),  # Turns off y-axis ticks
    )
)
plot6

dotplot¶

I was very disappointed with dotplot. The y axis seems totally broken - it looks like we get a normalized count (maximum = 1), even though we explicitly ask for a "count" (and even that is wrong)

In [11]:

plot7 = (
    p9.ggplot(data=df)
    + p9.geom_dotplot(
        mapping=p9.aes(
            x="x",
            fill="fill",
            y=p9.after_stat("count"),
        ),
        bins=40,
        dotsize=0.25,
    )
    + p9.scale_fill_identity()
    + p9.theme_bw()
    + p9.labs(subtitle="dotplot")
)
plot7

Flip axis to get a vertical graphic fails completely! I wont be using this in the final graphic.

In [12]:

plot8 = plot7 + p9.coord_flip()
plot8

jitter¶

In this graphic, we spread the points out to reduce overlap, and plot with a high transparency (low alpha). The y axis ticks and label aren't neccessary so they are removed

In [13]:

plot9 = (
    p9.ggplot(data=df)
    + p9.geom_jitter(
        mapping=p9.aes(x="x", y=0, fill="fill"),
        width=0,
        height=0.25,
        alpha=0.2,
    )
    + p9.scale_fill_identity()
    + p9.theme_bw()
    + p9.labs(subtitle="jitter")
    + p9.ylim((-1, 1))
)
plot9 + p9.theme(
    axis_title_y=p9.element_blank(),  # Removes the y-axis tick labels
    axis_text_y=p9.element_blank(),  # Removes the y-axis tick labels
    axis_ticks_y=p9.element_blank(),  # Turns off y-axis ticks
    # panel_grid=p9.element_blank(),
)

Flip axis to get a horizontal graphic. We adjust the labels on what is now the x axis

In [14]:

plot10 = plot9 + p9.coord_flip()
plot10 + p9.theme(
    axis_title_x=p9.element_blank(),  # Removes the y-axis tick labels
    axis_text_x=p9.element_blank(),  # Removes the y-axis tick labels
    axis_ticks_x=p9.element_blank(),  # Turns off y-axis ticks
    # panel_grid=p9.element_blank(),
)

density¶

density produces a density curve. In this, and the two graphics below, we wdd the jittered points for comparison. One point to note: because we haven't declared "white" to be a color by the + p9.scale_fill_identity() call, it gets treated by geom_jitter as a categorical variable, and is allocated the first color in the color map (which happend to be red). That is why the jittered points have a pink tinge (alpha = 0.05, i.e. very transparent)

In [15]:

plot11 = (
    p9.ggplot(data=df)
    + p9.geom_density(mapping=p9.aes(x=x, y=p9.stage(start=0, after_stat="scaled")))
    + p9.theme_bw()
    + p9.labs(subtitle="density\n+ jitter")
    + p9.geom_jitter(
        mapping=p9.aes(x="x", y=0, fill="fill"),
        width=0,
        height=0.05,
        alpha=0.05,
        show_legend=False,
    )
)
plot11

Reset the jittered points to be white filled

In [16]:

plot11 + p9.scale_fill_identity()

Flip axis to get a vertical graphic

In [17]:

plot12 = plot11 + p9.coord_flip()
plot12

violin¶

Violin plots are another way of looking at density curves. They default to being drawn as vertical curves

In [18]:

plot13 = (
    p9.ggplot(data=df)
    + p9.geom_violin(
        mapping=p9.aes(y=x),
    )
    + p9.geom_jitter(
        mapping=p9.aes(x=0, y=x, fill="fill"),
        width=0.05,
        height=0,
        alpha=0.05,
        show_legend=False,
    )
    + p9.theme_bw()
    + p9.labs(subtitle="violin\n+ jitter")
)
plot13

Flip axis to get a horizontal graphic (not used subsequently)

In [19]:

plot14 = plot13 + p9.coord_flip()
plot14

boxplot¶

I am not a big fan of boxplots. They are hard to understand, and as can be seen below, they can give a completely incorrect impression as to density distribution

In [20]:

plot15 = (
    p9.ggplot(data=df)
    + p9.geom_boxplot(
        mapping=p9.aes(y=x, x=0),
    )
    + p9.geom_jitter(
        mapping=p9.aes(x=0, y=x, fill="fill"),
        width=0.05,
        height=0,
        alpha=0.05,
        show_legend=False,
    )
    + p9.theme_bw()
    + p9.labs(subtitle="boxplot\n+ jitter")
)
plot15

Flip axis to get a horizontal graphic (not used subsequently)

In [21]:

plot16 = plot15 + p9.coord_flip()
plot16

sina (again)¶

Because we will be compressing plots in the x direction, we reduce the default transparency of the random poinrs, and their size

In [22]:

plot17 = (
    p9.ggplot(data=df)
    + p9.geom_sina(
        mapping=p9.aes(y=x, x=0),
        alpha=0.2,
        size=0.2,
    )
    + p9.theme_bw()
    + p9.labs(subtitle="sina")
)
plot17

$No description has been provided for this image$

Final composite graphic¶

First of all we tidy up the y axis titles (where not needed)

In [23]:

plot4 = plot4 + p9.theme(
    axis_title_y=p9.element_blank(),
)  # Removes the y-axis title)

plot10 = plot10 + p9.theme(
    axis_title_y=p9.element_blank(),
)  # Removes the y-axis title)

plot12 = plot12 + p9.theme(
    axis_title_y=p9.element_blank(),
)  # Removes the y-axis title)

plot17 = plot17 + p9.theme(
    axis_title_x=p9.element_blank(),
)  # Removes the y-axis title)

We put a title on the left-most plot, which will act as the overall title. Titles for compositions are coming to plotnine, but not just yet.

We build the composition, and tidy up the x axis, which we want unadorned

In [24]:

plot2 = (
    plot2
    + p9.labs(title="1D density visualization")
    + p9.theme(plot_title_position="plot")
)

(plot2 | plot4 | plot10 | plot12 | plot13 | plot15 | plot17) * p9.theme(
    figure_size=(10, 6),
    axis_title_x=p9.element_blank(),  # Removes the x-axis title
    axis_text_x=p9.element_blank(),  # Removes the x-axis tick labels
    axis_ticks_x=p9.element_blank(),  # Turns off x-axis ticks
    # panel_grid=p9.element_blank(),
)

Conclusions¶

This demonstrates the minimal code needed for a quiet nice graphic. Plotnine defaults are really very well chosen in my experience.

Also, steer clear of dotplot, and refrain from boxplots

Reproducability¶

In [25]:

%watermark

Last updated: 2026-03-09T16:15:37.002049+10:00

Python implementation: CPython
Python version       : 3.11.14
IPython version      : 9.10.0

Compiler    : MSC v.1929 64 bit (AMD64)
OS          : Windows
Release     : 10
Machine     : AMD64
Processor   : Intel64 Family 6 Model 170 Stepping 4, GenuineIntel
CPU cores   : 22
Architecture: 64bit

In [26]:

%watermark -h -iv -co

conda environment: fun_minim

Hostname: INSPIRON16

numpy   : 2.4.1
pandas  : 2.3.3
plotnine: 0.15.0

In [27]:

import contextlib

import ipynbname

with contextlib.suppress(FileNotFoundError):
    print(f"Notebook file name: {ipynbname.name()}")
# end with

Notebook file name: one_d_density

In [ ]:

Plotnine - a comparison of 1D density graphs

Plotnine and 1D density¶

Introduction¶

Implementation¶

Imports¶

Magics¶

Create dataset¶

Graphics¶

Point plot¶

Histogram¶

sina¶

dotplot¶

jitter¶

density¶

violin¶

boxplot¶

sina (again)¶

Final composite graphic¶

Conclusions¶

Reproducability¶

Comments

About