import pandas as pd
from plotnine import (
ggplot,
aes,
geom_col,
geom_text,
position_dodge,
lims,
theme,
element_text,
element_blank,
element_rect,
element_line, )
Two Variable Bar Plot
Visualising on a single plot the values of a variable that has nested (and independent) variables
Create the data
= pd.DataFrame(
df
{"variable": [
"gender",
"gender",
"age",
"age",
"age",
"income",
"income",
"income",
"income",
],"category": [
"Female",
"Male",
"1-24",
"25-54",
"55+",
"Lo",
"Lo-Med",
"Med",
"High",
],"value": [60, 40, 50, 30, 20, 10, 25, 25, 40],
}
)"variable"] = pd.Categorical(df["variable"], categories=["gender", "age", "income"])
df["category"] = pd.Categorical(df["category"], categories=df["category"])
df[
df
variable | category | value | |
---|---|---|---|
0 | gender | Female | 60 |
1 | gender | Male | 40 |
2 | age | 1-24 | 50 |
3 | age | 25-54 | 30 |
4 | age | 55+ | 20 |
5 | income | Lo | 10 |
6 | income | Lo-Med | 25 |
7 | income | Med | 25 |
8 | income | High | 40 |
We want to visualise this data and at a galance get an idea to how the value
breaks down along the category
s for the different variable
. Note that each variable
has different category
s.
First we make a simple plot with all this information and see what to draw from it.
(="variable", y="value", fill="category"))
ggplot(df, aes(x+ geom_col()
)
All the value
s along each variable add up to 100, but stacked together the difference within and without the groups is not clear. The solution is to dodge
the bars.
(="variable", y="value", fill="category"))
ggplot(df, aes(x+ geom_col(stat="identity", position="dodge") # modified
)
This is good, it gives us the plot we want but the legend is not great. Each variable
has a different set of category
s, but the legend has them all clamped together. We cannot easily change the legend, but we can replicate it’s purpose by labelling the individual bars.
To do this, we create a geom_text
with position_dodge(width=0.9)
to match the ratio of the space taken up by each variable. If there was no spacing between the bars of different variables, we would have width=1
.
A minor quack, when text extends beyond the limits we have to manually make space or it would get clipped. Therefore we adjust the bottom y
limits.
= position_dodge(width=0.9) # new
dodge_text
(="variable", y="value", fill="category"))
ggplot(df, aes(x+ geom_col(stat="identity", position="dodge", show_legend=False) # modified
+ geom_text(
=-0.5, label="category"), # new
aes(y=dodge_text,
position="gray",
color=8,
size=45,
angle="top",
va
)+ lims(y=(-5, 60)) # new
)
Would it look too crowded if we add value labels on top of the bars?
= position_dodge(width=0.9)
dodge_text
(="variable", y="value", fill="category"))
ggplot(df, aes(x+ geom_col(stat="identity", position="dodge", show_legend=False)
+ geom_text(
=-0.5, label="category"),
aes(y=dodge_text,
position="gray",
color=8,
size=45,
angle="top",
va
)+ geom_text(
="value"), # new
aes(label=dodge_text,
position=8,
size="bottom",
va="{}%",
format_string
)+ lims(y=(-5, 60))
)
That looks okay. The value
s line up with the category
s because we used the same dodge
parameters. For the final polish, we remove the y-axis, clear out the panel and make the variable
and category
labels have the same color.
# Gallery, bars
= position_dodge(width=0.9)
dodge_text = "#555555"
ccolor
(="variable", y="value", fill="category"))
ggplot(df, aes(x+ geom_col(stat="identity", position="dodge", show_legend=False)
+ geom_text(
=-0.5, label="category"),
aes(y=dodge_text,
position=ccolor,
color=8,
size=45,
angle="top",
va# modified
) + geom_text(
="value"),
aes(label=dodge_text,
position=8,
size="bottom",
va="{}%",
format_string
)+ lims(y=(-5, 60))
+ theme(
=element_rect(fill="white"), # new
panel_background=element_blank(),
axis_title_y=element_line(color="black"),
axis_line_x=element_blank(),
axis_line_y=element_blank(),
axis_text_y=element_text(color=ccolor),
axis_text_x=element_blank(),
axis_ticks_major_y=element_blank(),
panel_grid=element_blank(),
panel_border
) )
Credit: I saved a plot this example is based on a while ago and forgot/misplaced the link to the source. The user considered it a minor coup.