Dumbell chart with custom annotations

A lollipop chart is a type of chart that is a variation of a barplot, where the bars are replaced by dots and lines. This example shows how to create a dumbell chart, which differs from a lollipop chart by having lines connecting the dots.

In this post, we will reproduce a complex version of a dumbell chart, with step-by-step explanations and code snippets.

About

This chart displays the percentage of wins and losses of football teams with a lollipop chart, originally created by Cédric Scherer in R. You can find the R version here.

This post is a reproduction of it in Python by Joseph Barbier.

As a teaser, here is the chart we will reproduce:

preview final chart

Libraries

First, we need to install the following libraries:

# libraries
import matplotlib.pyplot as plt
from highlight_text import ax_text, fig_text
import pandas as pd

Dataset

We start by manually creating the dataset and put it in a pandas dataframe.

Then we just need to to divide the number of wins and losses by the total number of games to get the win/loss rate.

df = pd.DataFrame({
    "team": ["FC Bayern München", "SV Werder Bremen", "Borussia Dortmund", "VfB Stuttgart", 
             "Borussia M'gladbach", "Hamburger SV", "Eintracht Frankfurt",
             "FC Schalke 04", "1. FC Köln", "Bayer 04 Leverkusen"],
    "matches": [2000, 1992, 1924, 1924, 1898, 1866, 1856, 1832, 1754, 1524],
    "won": [1206, 818, 881, 782, 763, 746, 683, 700, 674, 669],
    "lost": [363, 676, 563, 673, 636, 625, 693, 669, 628, 447]
})

df['won'] = df['won'] / df['matches']
df['lost'] = df['lost'] / df['matches']
df.sort_values(by='matches', inplace=True)
df.reset_index(drop=True, inplace=True)
df.head()

	team	matches	won	lost
0	Bayer 04 Leverkusen	1524	0.438976	0.293307
1	1. FC Köln	1754	0.384265	0.358039
2	FC Schalke 04	1832	0.382096	0.365175
3	Eintracht Frankfurt	1856	0.367996	0.373384
4	Hamburger SV	1866	0.399786	0.334941

Most simple lollipop chart

Since there is no direct way to create a lollipop chart in matplotlib, we need to create it ourselves.

The code will simply plot individual lines for each team using the hlines() function and then plot data points on top of it using the scatter() function.
The hlines() function needs the y-coordinates (which simply is 0,1,2,...,n, where n is the number of team minus 1) and the x-coordinates (which is the win/loss rate).
The zorder parameter is used to make sure the data points are plotted on top of the lines.

fig, ax = plt.subplots(figsize=(8,5))

# horizontal lines
my_range = range(df['team'].nunique())
ax.hlines(y=my_range, xmin=df['lost'], xmax=df['won'], color='grey', alpha=0.4)

# points
ax.scatter(df['lost'], my_range, zorder=2)
ax.scatter(df['won'], my_range, zorder=2)

plt.show()

Custom style

Let's add the following features to the chart:

define color_lose and color_win to color the lollipop chart
change the size of the data points with the s argument
remove the axis with ax.set_axis_off()

color_lose, color_win = "#EFAC00", "#28A87D"

fig, ax = plt.subplots(figsize=(8,5))

# horizontal lines
my_range = range(df['team'].nunique())
ax.hlines(y=my_range, xmin=df['lost'], xmax=df['won'], color='grey', alpha=0.4)

# points
ax.scatter(df['lost'], my_range, color=color_lose, zorder=2, s=80)
ax.scatter(df['won'], my_range, color=color_win, zorder=2, s=80)

# remove axis
ax.set_axis_off()

plt.show()

Add team names and values

Now we want to add the team names and the percentage of wins and losses to the chart.

In practice we iterate over the teams with a for loop and add the text with the ax_text() function.
Text formatting such as colors and boldness is done thanks to the highlight_text package

color_lose, color_win = "#EFAC00", "#28A87D"

fig, ax = plt.subplots(figsize=(7,5))

# horizontal lines
my_range = range(df['team'].nunique())
ax.hlines(y=my_range, xmin=df['lost'], xmax=df['won'], color='grey', alpha=0.4)

# points
ax.scatter(df['lost'], my_range, color=color_lose, zorder=2, s=80)
ax.scatter(df['won'], my_range, color=color_win, zorder=2, s=80)

# add team names
n = len(df)
for i in range(df['team'].nunique()):

    # losses
    losses = df['lost'][i]
    ax_text(
        losses-0.012, i,
        f"<{df['lost'][i]*100:.0f}%>",
        ha='right', va='center',
        highlight_textprops=[
            {"color": color_lose,
             "weight": "light"}]
    )

    # wins
    wins = df['won'][i]
    ax_text(
        wins+0.012, i,
        f"<{df['won'][i]*100:.0f}%>",
        ha='left', va='center',
        highlight_textprops=[
            {"color": color_win,
             "weight": "light"}]
    )

    # team names
    team_name = df['team'][i]
    ax_text(0.1, i,
            f"<{team_name}>",
            ha='right', va='center',
            highlight_textprops=[
                {"color": "black",
                 "weight": "light",
                 "size": 12}]
        )
    
# remove axis
ax.set_axis_off()

plt.tight_layout()
plt.show()

Colors details

In the original post, there is not only yellow and green, but also a darker version of these colors. Even if this may sound like a detail, it is actually an important feature of the chart since it makes it easier to read values.

With that, we take the opportunity to define a special case (with a if statement during the loop) for the FC Bayern München team that we want to highlight. We add a darker horizontal line and a bold team name.

color_lose, color_win = "#EFAC00", "#28A87D"
color_lose_dark, color_win_dark = "#aa7c05", "#1e8563"

fig, ax = plt.subplots(figsize=(7,5))

# horizontal lines
my_range = range(df['team'].nunique())
ax.hlines(y=my_range, xmin=df['lost'], xmax=df['won'], color='grey', alpha=0.4)

# points
ax.scatter(df['lost'], my_range, color=color_lose, zorder=3, s=80)
ax.scatter(df['won'], my_range, color=color_win, zorder=2, s=80)

# add team names
n = len(df)
for i in range(df['team'].nunique()):
    
    # team names
    team_name = df['team'][i]
    if team_name == "FC Bayern München":
        weight = "bold"
        bayern_munch = df[df['team'] == team_name]
        ax.hlines(y=i, xmin=bayern_munch['lost'], xmax=bayern_munch['won'], linewidth=2, color='black', zorder=1)

    else:
        weight = "normal"
    ax_text(0.1, i,
            f"<{team_name}>",
            ha='right', va='center',
            highlight_textprops=[
                {"color": "black",
                 "fontweight": weight,
                 "size": 12}
            ]
        )

    # losses
    losses = df['lost'][i]
    ax_text(
        losses-0.016, i,
        f"<{df['lost'][i]*100:.0f}%>",
        ha='right', va='center',
        highlight_textprops=[
            {"color": color_lose_dark,
             "weight": weight}
        ]
    )

    # wins
    wins = df['won'][i]
    ax_text(
        wins+0.016, i,
        f"<{df['won'][i]*100:.0f}%>",
        ha='left', va='center',
        highlight_textprops=[
            {"color": color_win_dark,
             "weight": weight}]
    )

# remove axis
ax.set_axis_off()

plt.tight_layout()
plt.show()

Title and annotations

Now that we have the core of the chart, we only need to add a few annotations. Since the original chart has some specific fonts, we have to define them before:

First we load the Familjen Grotesk font:

# find path to font
from matplotlib import font_manager
for fontpath in font_manager.findSystemFonts(fontpaths=None, fontext='ttf'):
    if 'grotesk' in fontpath.lower():
        print(fontpath)

# get font properties
from matplotlib.font_manager import FontProperties
personal_path = '/Users/josephbarbier/Library/Fonts/'

# font
font_path = personal_path + 'FamiljenGrotesk-VariableFont_wght.ttf'
grotesk_font = FontProperties(fname=font_path)

# bold font
font_path_bold = personal_path + 'FamiljenGrotesk-Bold.ttf'
grotesk_font_bold = FontProperties(fname=font_path_bold)

/Users/josephbarbier/Library/Fonts/FamiljenGrotesk-VariableFont_wght.ttf
/Users/josephbarbier/Library/Fonts/FamiljenGrotesk-SemiBold.ttf
/Users/josephbarbier/Library/Fonts/FamiljenGrotesk-Bold.ttf

Then we load the Pally font:

# find path to font
from matplotlib import font_manager
for fontpath in font_manager.findSystemFonts(fontpaths=None, fontext='ttf'):
    if 'pally' in fontpath.lower():
        print(fontpath)

# get font properties
from matplotlib.font_manager import FontProperties
personal_path = '/Users/josephbarbier/Library/Fonts/'

# font
font_path = personal_path + 'Pally-Regular.otf'
pally_font = FontProperties(fname=font_path)

# bold font
font_path_bold = personal_path + 'Pally-Bold.otf'
pally_font_bold = FontProperties(fname=font_path_bold)

/Users/josephbarbier/Library/Fonts/Pally-Regular.otf
/Users/josephbarbier/Library/Fonts/Pally-Bold.otf
/Users/josephbarbier/Library/Fonts/Pally-Medium.otf

Other annotations are added with the ax_text() function. The positionings are defined through trial and error until it looks good.

color_lose, color_win = "#EFAC00", "#28A87D"
color_lose_dark, color_win_dark = "#aa7c05", "#1e8563"

fig, ax = plt.subplots(figsize=(7,5))

# horizontal lines
my_range = range(df['team'].nunique())
ax.hlines(y=my_range, xmin=df['lost'], xmax=df['won'], color='grey', alpha=0.4)

# points
ax.scatter(df['lost'], my_range, color=color_lose, zorder=2, s=80)
ax.scatter(df['won'], my_range, color=color_win, zorder=2, s=80)

# add team names
n = len(df)
for i in range(df['team'].nunique()):
    
    # team names
    team_name = df['team'][i]
    if team_name == "FC Bayern München":
        font = grotesk_font_bold
        bayern_munch = df[df['team'] == team_name]
        ax.hlines(y=i, xmin=bayern_munch['lost'], xmax=bayern_munch['won'], linewidth=2, color='black', zorder=1)

    else:
        font = grotesk_font
    ax_text(0.1, i,
            f"<{team_name}>",
            ha='right', va='center',
            fontproperties=grotesk_font,
            highlight_textprops=[
                {"color": "black",
                 "font": font,
                 "size": 12}
            ]
        )

    # losses
    losses = df['lost'][i]
    ax_text(
        losses-0.016, i,
        f"<{df['lost'][i]*100:.0f}%>",
        ha='right', va='center',
        fontproperties=grotesk_font,
        highlight_textprops=[
            {"color": color_lose_dark,
             "font": font}
        ]
    )

    # wins
    wins = df['won'][i]
    ax_text(
        wins+0.016, i,
        f"<{df['won'][i]*100:.0f}%>",
        ha='left', va='center',
        fontproperties=grotesk_font,
        highlight_textprops=[
            {"color": color_win_dark,
             "font": font}]
    )

# title
text = "Among the ten teams with the most Bundesliga matches,\n<Bayern Munich has the> <lowest loss> <and> <highest win> <rates>"
fig_text(
    0.05, 1,
    text,
    fontsize=20,
    fontproperties=grotesk_font,
    ha='left', va='center',
    highlight_textprops=[
        {"font": grotesk_font_bold},
        {"color": color_lose_dark,
         "font": grotesk_font_bold},
        {"font": grotesk_font_bold},
        {"color": color_win_dark,
         "font": grotesk_font_bold},
        {"font": grotesk_font_bold},
    ]
)

# legend annotation
text = '<matches lost>'
fig_text(
    0.38, 0.8,
    text,
    fontsize=13,
    ha='center', va='center',
    highlight_textprops=[
        {"color": color_lose,
         "font": pally_font_bold}
    ]
)
text = '<matches won>'
fig_text(
    0.76, 0.8,
    text,
    fontsize=13,
    ha='center', va='center',
    highlight_textprops=[
        {"color": color_win,
         "font": pally_font_bold}
    ]
)

# credit
text = "<Graphic: Cédric Scherer · Source: kicker.de ·> Teams are ranked according to the number of Bundesliga matches they have played so far"
fig_text(
    0.05, 0.04,
    text,
    fontsize=8,
    fontproperties=grotesk_font,
    color='grey',
    ha='left', va='center',
    highlight_textprops=[
        {"font": grotesk_font_bold}
    ]
)

# remove axis
ax.set_axis_off()

# note about Frankfurt
text = "<Frankfurt> is the only team listed\nthat <lost more games (37.3%)>\nthan <they have won (36.8%)>"
fig_text(
    0.75, 0.2,
    text,
    fontsize=12,
    color='grey',
    fontproperties=grotesk_font,
    ha='left', va='center',
    highlight_textprops=[
        {"font": pally_font_bold},
        {"color": color_lose_dark,
         "font": pally_font_bold},
        {"color": color_win_dark,
         "font": pally_font_bold}
    ]
)

# arrow
from matplotlib.patches import FancyArrowPatch
arrow_style = "Simple, tail_width=0.5, head_width=4, head_length=8"
connection_style = "arc3,rad=.3"
arrow_properties = {
    "arrowstyle": arrow_style,
    "color": "grey",
}
tail_position = (0.62, 1.6)
head_position = (0.43, 3)
arrow = FancyArrowPatch(
    tail_position, head_position,
    connectionstyle=connection_style,
    **arrow_properties
)
ax.add_patch(arrow)

plt.tight_layout()
plt.show()

Going further

This article explains how to create a lollipop chart in matplotlib.

You might also be interested in this other beautiful dumbell chart and more generally by how to create beautiful annotations in matplotlib

Barplot

Spider / Radar

Wordcloud

Parallel

Lollipop

Circular Barplot

Table

🚨 Grab the Data To Viz poster!

Do you know all the chart types? Do you know which one you should pick? I made a decision tree that answers those questions. You can download it for free!