Lecture 9 | Grammar of Graphics II

IS 616: Large Scale Data Analysis and Visualization

text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text text

https://r-graph-gallery.com/

https://github.com/holtzy/R-graph-gallery

https://python-graph-gallery.com/

https://github.com/holtzy/The-Python-Graph-Gallery

Using the “data-to-viz classification”

https://www.data-to-viz.com/

https://en.wikipedia.org/wiki/R/dataisbeautiful

https://www.reddit.com/r/dataisbeautiful/

import datetime

import pandas as pd
from plotnine import *

“Visualizing Baby Sleep Times in Python”

All code from: https://github.com/dodger487/snoo_plots

For more analyses of the same data that we won’t cover: https://www.relevantmisc.com/python/r/data/2020/06/10/baby-sleep-night-day/

df = pd.read_csv(
  'https://github.com/\
dodger487/snoo_plots/\
raw/master/sleep_data.csv'
  )

df.head()

            start_time             end_time  duration  asleep  soothing
0  2019-11-21 18:30:32  2019-11-21 18:31:52        80      80         0
1  2019-11-22 04:03:22  2019-11-22 05:07:17      3835    2422      1413
2  2019-11-22 05:53:24  2019-11-22 05:54:27        63      36        27
3  2019-11-22 05:52:51  2019-11-22 07:01:11      4100    3140       960
4  2019-11-22 22:51:19  2019-11-22 23:20:48      1769    1289       480

We want to plot date on the x-axis and the time of day on the y-axis. We’ll have a vertical line from the start to the end of a sleep session.

The datetimes include both dates and times so we’ll have to break them apart. Pandas .dt is great here, it allows you to use the standard library datetime methods on a Pandas series in parallel.

# Break out dates and times.
df["start_datetime"]\
= pd.to_datetime(\
df["start_time"])

df["end_datetime"]\
= pd.to_datetime(\
df["end_time"])

df["start_time"]\
= df["start_datetime"].dt.time

df["end_time"]\
= df["end_datetime"].dt.time

df["start_date"]\
= df["start_datetime"].dt.date

What about sleep sessions that span days?

# Deal with sessions that
# cross day boundaries.

df_no_cross =\
df[df["start_datetime"].dt.day\
== df["end_datetime"].dt.day].copy()

df_cross =\
df[df["start_datetime"].dt.day\
!= df["end_datetime"].dt.day]

df_cross_1 = df_cross.copy()

df_cross_2 = df_cross.copy()

First, we separate out sessions into those that cross midnight and those that don’t. We don’t need to do anything about the former and can set that data aside.

For sessions that do cross midnight, we make two copies.

df_cross_1["end_time"]\
= datetime.time(\
hour=23, minute=59, second=59)

df_cross_2["start_date"]\
= df_cross_2["start_date"]\
+ datetime.timedelta(days=1)

df_cross_2["start_time"]\
= datetime.time(\
hour=0, minute=0, second=0)

For the first copy, we set the end date to just before midnight.

For the second, we set the start time to midnight and increment the date to be the next day.

# Combine dataframes

rows_no_cross =\
df_no_cross[["start_date",
"start_time", "end_time"]]

rows_cross_1 =\
df_cross_1[["start_date",
"start_time", "end_time"]]

rows_cross_2 =\
df_cross_2[["start_date",
"start_time", "end_time"]]

rows =\
pd.concat([rows_no_cross,
rows_cross_1,
rows_cross_2])

Finally, we combine these dataframes into one new dataframe, which we’ll use for plotting.

# Convert back to datetime
# so plotnine can understand it

rows["start_time"] =\
pd.to_datetime(rows["start_time"],
format='%H:%M:%S')

rows["end_time"] =\
pd.to_datetime(rows["end_time"],
format='%H:%M:%S')

We will use plotnine in Python to make the visualization using the ggplot2 syntax.

plot = (ggplot(data=rows)
  + aes(x="start_date")
  + geom_linerange(aes(
    ymin = "start_time",
    ymax = "end_time"))
  + scale_x_date(name="",
    date_labels="%b",
    expand=(0, 0)) 
  + scale_y_datetime(
    date_labels="%H:%M",
    expand=(0, 0, 0, 0.0001))
  + ggtitle(
    "Baby Sleep Times")
  + theme_minimal() 
  + theme(
    plot_background=\
    element_rect(
      color="white")))

There’s clearly some missing data: “We didn’t track lots of naps, and a few days have no data at all.”

“That said, some clear patterns emerge. During the first few days, sleep is all over the place, but gradually settles into a routine. Later, nighttime wake-ups become fewer and shorter.” (https://www. relevantmisc.com/r/python/2020/05/26/visualizing-baby-sleep/)

What makes the baby sleep data interesting?

https://www.washingtonpost.com/news/wonk/wp/2017/01/05/what-its-like-to-sleep-like-a-baby-visualized-by-a-dad/

https://www.reddit.com/r/dataisbeautiful/comments/ 5l39mu/my_daughters_sleeping_patterns_for_the_first_4/

Short switch to R…

No coord_polar in plotnine

https://github.com/has2k1/plotnine/issues/10

Maybe at some point?

Make yourself flexible

Using R in Jupyter Notebooks

https://docs.anaconda.com/free/navigator/tutorials/r-lang/

Using Python in Rmarkdown code chunks

library(ggplot2)
library(dplyr)
library(readr)

l1 <- 'https://github.com/dodger487/'
l2 <- 'snoo_plots/raw/master/sleep_data.csv'

df <- read_csv(paste0(l1,l2))

# We need to add rows for when baby is awake
# and do the inverse when the baby is asleep

df <- df %>%
  select(-duration, -asleep, -soothing) %>%
  mutate(session_type = "asleep")

Credits: https://www.relevantmisc.com/r/2020/06/01/baby-sleep-radial/

inverse_df <- df %>%
  arrange(start_time) %>%
  mutate(
    start_time_new = end_time,
    end_time_new = lead(start_time),
    session_type = "awake",
    start_time = start_time_new,
    end_time = end_time_new
  ) %>%
  select(-start_time_new, -end_time_new) %>%
  filter(!is.na(start_time) & !is.na(end_time))

# Combine the "awake" and "asleep" rows

df <- rbind(df, inverse_df) %>%
arrange(start_time)

Again, we need to break up sessions that cross the midnight boundary into two sessions, one pre-midnight and one-after midnight, so that all sessions only take place in one day.

df_no_cross <- df %>% 
  filter(date(start_time) == date(end_time)) %>%
  mutate(
    start_date = date(start_time), 
    next_date = start_date + days(1),
    start_time = hms::as_hms(start_time),
    end_time = hms::as_hms(end_time))

df_cross <- df %>% filter(date(start_time) != date(end_time))

df_cross_1 <- df_cross %>% 
  mutate(
    start_date = date(start_time), 
    next_date = start_date + days(1),
    start_time = hms::as_hms(start_time),
    end_time = hms::as_hms("23:59:59")
  )

We’ll simply break any row that crosses midnight into 2 sessions, one that ends at midnight and one that starts at midnight (as previously in Python, now in R).

df_cross_2 <- df_cross %>% 
  mutate(
    start_date = date(end_time), 
    next_date = start_date + days(1),
    start_time = hms::as_hms("00:00:00"),
    end_time = hms::as_hms(end_time)
  )

# Combine dataframes

rows <- rbind(
  df_no_cross,
  df_cross_1,
  df_cross_2
)

Now on to the visualization! We can use much of the code from plotnine before right away.

rows %>%
  ggplot(.) +
  aes(xmin=start_time,
  xmax=end_time,
  ymin=start_date,
  ymax=next_date,
  fill=session_type) +
  geom_rect() +
  facet_wrap(~session_type)

First, let’s look at this Cartesian axis plot that faceted by awake and asleep to check if everything looks OK.

rows %>%
  ggplot(.) +
  aes(xmin=start_time,
  xmax=end_time,
  ymin=start_date,
  ymax=next_date,
  fill=session_type) +
  geom_rect() #+
  # facet_wrap(~session_type)

p <- (rows %>%
    filter(session_type == "asleep") %>%
    ggplot(aes(x=start_date), data=.)
  + geom_linerange(aes(ymin = start_time, ymax = end_time))
  + scale_x_date(name="", date_labels="%b", expand=c(0, 0)) 
  + scale_y_time(labels = function(x)
    format(as.POSIXct(x),format = '%H:%M'),
                 expand=c(0, 0, 0, 0.0001))
  + ggtitle("Baby Sleep Times")
  + theme_minimal())

p + coord_polar(start=0)

It appears that our axes are flipped: We want the time of day to be the angle, and the radius to be the day.

p + coord_polar(start = 0, theta = "y")

Not bad! Let’s add colors and create the final plot!

# Create custom colors, pulled from original plot
color_awake <- rgb(248/256, 205/256, 160/256)
color_sleep <- rgb(63/256, 89/256, 123/256)

# Create radial plot
rows %>%
  filter(start_date <= "2020-05-20") %>%
  ggplot(aes(x=start_date), data=.) +
    geom_linerange(aes(ymin = start_time,
                       ymax = end_time,
                       color = session_type)) +
  scale_x_date(name="", date_labels="%b", expand=c(0, 28)) +
  scale_y_time(expand=c(0, 0, 0, 0.0001)) +
  scale_color_manual(values = c(color_sleep, color_awake)) +
  theme_void() +
  coord_polar(theta = "y") +
  theme(legend.position = "none")

“The Vienna school, on the other hand, postulates: to remember simplified pictures is better than to forget accurate figures.” (Neurath, 1973; p. 220)

Neurath, O. (1973). Empiricism and Sociology (M. Neurath & R. S. Cohen, Eds.). Springer Netherlands. https://doi.org/10.1007/978-94-010-2525-6

A symbol should

“Follow principles of good design.
Be usable in either large or small sizes.
Represent a general concept, not an individual one.
Be clearly distinguishable from other symbols.
Be interesting.
Be capable of being used as a counting unit.
Be usable in outline or in silhouette.”

https://en.wikipedia.org/wiki/Rudolf_Modley

Isotype as chartjunk?

Isotype is not prone to the distortions measured with the Lie Factor, when certain principles are taken into account:

“The first rule of Isotype is that greater quantities are not represented by an enlarged pictogram but by a greater number of the same-sized pictogram.

In Neurath’s view, variation in size does not allow accurate comparison (what is to be compared – height/length or area?) whereas repeated pictograms, which always represent a fixed value within a certain chart, can be counted if necessary.

Isotype pictograms almost never depicted things in perspective in order to preserve this clarity, and there were other guidelines for graphic configuration and use of colour.”

https://en.wikipedia.org/wiki/Isotype_(picture_language)

Isotype in the context of this course

So far, we modified:

Scales, for example with the logarithm for gdppercapita

The coordinate system with coord_polar for example

And now we will take a look at geoms and try to emulate ISOTYPE style; as it will turn out, we can pretty easily choose arbitrary symbols to be mapped to our data (that’s also often the starting point of how new ggplot2 extensions get developed)

We heard already earlier about evidence that even “unnecessary” purely ornamental chartjunk helps to remember

If we add meaningful symbols directly to graphs, we make visualizations more immediate for the viewer and more self explanatory, because the geometric objects themselves serve as a legend

For more information and experiments on the perception of Isotype graphics, check out http://steveharoz.com/research/isotype/

“The first concept to understand is that Isotypes often mix qualitative and quantitative data.

By simplifying the concepts trying to be communicated (often qualitative) and then elaborating with pictograms (quantitative), Isotypes aggregate both types of information into an easy-to-understand message.”

https://nightingaledvs.com/lessons-of-isotype-part-1-only-an-ocean-between/

Our example comes from a series of books that promoted cultural understanding between Britain and its allies during World War II

import altair as alt
import pandas as pd

source = pd.DataFrame([
      {'country': 'Great Britain', 'animal': 'cattle'},
      {'country': 'Great Britain', 'animal': 'cattle'},
      {'country': 'Great Britain', 'animal': 'cattle'},
      {'country': 'Great Britain', 'animal': 'pigs'},
      {'country': 'Great Britain', 'animal': 'pigs'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'Great Britain', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'cattle'},
      {'country': 'United States', 'animal': 'pigs'},
      {'country': 'United States', 'animal': 'pigs'},
      {'country': 'United States', 'animal': 'pigs'},
      {'country': 'United States', 'animal': 'pigs'},
      {'country': 'United States', 'animal': 'pigs'},
      {'country': 'United States', 'animal': 'pigs'},
      {'country': 'United States', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'sheep'},
      {'country': 'United States', 'animal': 'sheep'}
    ])

domains = ['person', 'cattle', 'pigs', 'sheep']

shape_scale = alt.Scale(
    domain=domains,
    range=[
        'M1.7 -1.7h-0.8c0.3 -0.2 0.6 -0.5 0.6 -0.9c0 -0.6 -0.4 -1 -1 -1c-0.6 0 -1 0.4 -1 1c0 0.4 0.2 0.7 0.6 0.9h-0.8c-0.4 0 -0.7 0.3 -0.7 0.6v1.9c0 0.3 0.3 0.6 0.6 0.6h0.2c0 0 0 0.1 0 0.1v1.9c0 0.3 0.2 0.6 0.3 0.6h1.3c0.2 0 0.3 -0.3 0.3 -0.6v-1.8c0 0 0 -0.1 0 -0.1h0.2c0.3 0 0.6 -0.3 0.6 -0.6v-2c0.2 -0.3 -0.1 -0.6 -0.4 -0.6z',
        'M4 -2c0 0 0.9 -0.7 1.1 -0.8c0.1 -0.1 -0.1 0.5 -0.3 0.7c-0.2 0.2 1.1 1.1 1.1 1.2c0 0.2 -0.2 0.8 -0.4 0.7c-0.1 0 -0.8 -0.3 -1.3 -0.2c-0.5 0.1 -1.3 1.6 -1.5 2c-0.3 0.4 -0.6 0.4 -0.6 0.4c0 0.1 0.3 1.7 0.4 1.8c0.1 0.1 -0.4 0.1 -0.5 0c0 0 -0.6 -1.9 -0.6 -1.9c-0.1 0 -0.3 -0.1 -0.3 -0.1c0 0.1 -0.5 1.4 -0.4 1.6c0.1 0.2 0.1 0.3 0.1 0.3c0 0 -0.4 0 -0.4 0c0 0 -0.2 -0.1 -0.1 -0.3c0 -0.2 0.3 -1.7 0.3 -1.7c0 0 -2.8 -0.9 -2.9 -0.8c-0.2 0.1 -0.4 0.6 -0.4 1c0 0.4 0.5 1.9 0.5 1.9l-0.5 0l-0.6 -2l0 -0.6c0 0 -1 0.8 -1 1c0 0.2 -0.2 1.3 -0.2 1.3c0 0 0.3 0.3 0.2 0.3c0 0 -0.5 0 -0.5 0c0 0 -0.2 -0.2 -0.1 -0.4c0 -0.1 0.2 -1.6 0.2 -1.6c0 0 0.5 -0.4 0.5 -0.5c0 -0.1 0 -2.7 -0.2 -2.7c-0.1 0 -0.4 2 -0.4 2c0 0 0 0.2 -0.2 0.5c-0.1 0.4 -0.2 1.1 -0.2 1.1c0 0 -0.2 -0.1 -0.2 -0.2c0 -0.1 -0.1 -0.7 0 -0.7c0.1 -0.1 0.3 -0.8 0.4 -1.4c0 -0.6 0.2 -1.3 0.4 -1.5c0.1 -0.2 0.6 -0.4 0.6 -0.4z',
        'M1.2 -2c0 0 0.7 0 1.2 0.5c0.5 0.5 0.4 0.6 0.5 0.6c0.1 0 0.7 0 0.8 0.1c0.1 0 0.2 0.2 0.2 0.2c0 0 -0.6 0.2 -0.6 0.3c0 0.1 0.4 0.9 0.6 0.9c0.1 0 0.6 0 0.6 0.1c0 0.1 0 0.7 -0.1 0.7c-0.1 0 -1.2 0.4 -1.5 0.5c-0.3 0.1 -1.1 0.5 -1.1 0.7c-0.1 0.2 0.4 1.2 0.4 1.2l-0.4 0c0 0 -0.4 -0.8 -0.4 -0.9c0 -0.1 -0.1 -0.3 -0.1 -0.3l-0.2 0l-0.5 1.3l-0.4 0c0 0 -0.1 -0.4 0 -0.6c0.1 -0.1 0.3 -0.6 0.3 -0.7c0 0 -0.8 0 -1.5 -0.1c-0.7 -0.1 -1.2 -0.3 -1.2 -0.2c0 0.1 -0.4 0.6 -0.5 0.6c0 0 0.3 0.9 0.3 0.9l-0.4 0c0 0 -0.4 -0.5 -0.4 -0.6c0 -0.1 -0.2 -0.6 -0.2 -0.5c0 0 -0.4 0.4 -0.6 0.4c-0.2 0.1 -0.4 0.1 -0.4 0.1c0 0 -0.1 0.6 -0.1 0.6l-0.5 0l0 -1c0 0 0.5 -0.4 0.5 -0.5c0 -0.1 -0.7 -1.2 -0.6 -1.4c0.1 -0.1 0.1 -1.1 0.1 -1.1c0 0 -0.2 0.1 -0.2 0.1c0 0 0 0.9 0 1c0 0.1 -0.2 0.3 -0.3 0.3c-0.1 0 0 -0.5 0 -0.9c0 -0.4 0 -0.4 0.2 -0.6c0.2 -0.2 0.6 -0.3 0.8 -0.8c0.3 -0.5 1 -0.6 1 -0.6z',
        'M-4.1 -0.5c0.2 0 0.2 0.2 0.5 0.2c0.3 0 0.3 -0.2 0.5 -0.2c0.2 0 0.2 0.2 0.4 0.2c0.2 0 0.2 -0.2 0.5 -0.2c0.2 0 0.2 0.2 0.4 0.2c0.2 0 0.2 -0.2 0.4 -0.2c0.1 0 0.2 0.2 0.4 0.1c0.2 0 0.2 -0.2 0.4 -0.3c0.1 0 0.1 -0.1 0.4 0c0.3 0 0.3 -0.4 0.6 -0.4c0.3 0 0.6 -0.3 0.7 -0.2c0.1 0.1 1.4 1 1.3 1.4c-0.1 0.4 -0.3 0.3 -0.4 0.3c-0.1 0 -0.5 -0.4 -0.7 -0.2c-0.3 0.2 -0.1 0.4 -0.2 0.6c-0.1 0.1 -0.2 0.2 -0.3 0.4c0 0.2 0.1 0.3 0 0.5c-0.1 0.2 -0.3 0.2 -0.3 0.5c0 0.3 -0.2 0.3 -0.3 0.6c-0.1 0.2 0 0.3 -0.1 0.5c-0.1 0.2 -0.1 0.2 -0.2 0.3c-0.1 0.1 0.3 1.1 0.3 1.1l-0.3 0c0 0 -0.3 -0.9 -0.3 -1c0 -0.1 -0.1 -0.2 -0.3 -0.2c-0.2 0 -0.3 0.1 -0.4 0.4c0 0.3 -0.2 0.8 -0.2 0.8l-0.3 0l0.3 -1c0 0 0.1 -0.6 -0.2 -0.5c-0.3 0.1 -0.2 -0.1 -0.4 -0.1c-0.2 -0.1 -0.3 0.1 -0.4 0c-0.2 -0.1 -0.3 0.1 -0.5 0c-0.2 -0.1 -0.1 0 -0.3 0.3c-0.2 0.3 -0.4 0.3 -0.4 0.3l0.2 1.1l-0.3 0l-0.2 -1.1c0 0 -0.4 -0.6 -0.5 -0.4c-0.1 0.3 -0.1 0.4 -0.3 0.4c-0.1 -0.1 -0.2 1.1 -0.2 1.1l-0.3 0l0.2 -1.1c0 0 -0.3 -0.1 -0.3 -0.5c0 -0.3 0.1 -0.5 0.1 -0.7c0.1 -0.2 -0.1 -1 -0.2 -1.1c-0.1 -0.2 -0.2 -0.8 -0.2 -0.8c0 0 -0.1 -0.5 0.4 -0.8z'
    ]
)

color_scale = alt.Scale(
    domain=domains,
    range=['rgb(162,160,152)',
    'rgb(194,81,64)',
    'rgb(93,93,93)',
    'rgb(91,131,149)']
)

alt.Chart(source).mark_point(filled=True, opacity=1, size=100).encode(
    alt.X('x:O').axis(None),
    alt.Y('animal:O').axis(None),
    alt.Row('country:N').header(title=''),
    alt.Shape('animal:N').legend(None).scale(shape_scale),
    alt.Color('animal:N').legend(None).scale(color_scale),
).transform_window(
    x='rank()',
    groupby=['country', 'animal']
).properties(
    width=550,
    height=140
)

“We’ve seen how Isotypes group and organize both qualitative and quantitative data [..] “Population and Live Stock” is a great example of how the design of the grouped icons helps to reveal the story despite groups that are very different sizes.

The focus of the chart is not the overall size of the populations, but a subtle insight to tell a more interesting story.

By breaking the larger US population into three equal rows, it helps to make a more natural comparison between all four rows.”

“UK population in 1939 was 47.5M. US population in 1939 was 130.9M. This chart simplifies these numbers to emphasize the 1:3 ratio. So you can see there are three times as many Americans who eat less sheep and more pigs and cows.”

https://nightingaledvs.com/lessons-of-isotype-part-1-only-an-ocean-between/

Let’s make our lives a little bit easier…

https://github.com/gyli/PyWaffle

https://pywaffle.readthedocs.io/en/latest/index.html

pip install pywaffle

Often, this is already all we need as in the case of our next example

import pandas as pd

dict_users = {
  'Regular': 62,
  'New': 20,
  'Churned': 16,
  'Suspended': 2
  }
df = pd.Series(dict_users)

from pywaffle import Waffle
import matplotlib.pyplot as plt

fig = plt.figure(
  FigureClass=Waffle,
  figsize=(5,5),
  values=dict_users,
  rows=10
  )
  
plt.show()

In PyWaffle, we can use the characters parameter to provide a list of Unicode characters of the same length as the number of categories

Instead, if we want to use the same symbol for all the categories making them differ only by color, we can pass in a string with that character, for example, characters = '❤️'

We can also use the icons parameter in the same way that accepts a list of strings representing Font Awesome icons:

https://fontawesome.com/

colors_list = ['slateblue',
'limegreen', 'red', 'grey']

fig = plt.figure(
  FigureClass=Waffle,
  figsize=(5,5*1.3),
  values=dict_users,
  rows=10,
  colors=colors_list,
  icons=['user','user-plus',
  'user-minus', 'user-clock'],
  font_size=22,
  icon_legend=True,
  legend={
  'bbox_to_anchor': (0.8, 0),
  'fontsize': 15,
  'frameon': False})
  
plt.title('User dynamics',
fontsize=25)
plt.show()

To do similiar things in R

https://github.com/clauswilke/ggtextures

library(ggtextures)
library(grid)
library(magick)

data <- tibble(count = c(5, 3, 6), animal = c("giraffe", "elephant", "horse"),
  image = list(
    image_read_svg("http://steveharoz.com/research/isotype/icons/giraffe.svg"),
    image_read_svg("http://steveharoz.com/research/isotype/icons/elephant.svg"),
    image_read_svg("http://steveharoz.com/research/isotype/icons/horse.svg")))

ggplot(data, aes(animal, count, image = image)) +
  geom_isotype_col() + theme_minimal()

Acknowledgements

https://modley-telefact-1939-1945.tumblr.com/

https://medium.com/nightingale/the-telefacts-of-life-rudolf-modleys-isotypes-in-american-newspapers-1938-1945-d5478faa5647

http://www.thomwhite.co.uk/?p=1303

https://github.com/clauswilke/ggtextures

https://archive.ph/2023.02.07-211651/https://towardsdatascience.com/2-efficient-ways-of-creating-fancy-pictogram-charts-in-python-8b77d361d500