Making slope graphs by combining charts/geoms in plotnine (aka graphing based on multiple columns)¶
If your data isn't in a tidy or long format, sometimes things can be a little more difficult when using plotnine. But don't worry, all things are still possible! Let's take a look.
Steps to a slope graph¶
Let's say we have some information about school test scores, and how they changed between two years.
import pandas as pd
from plotnine import *
df = pd.DataFrame([
{ 'school_name': 'School A', '2015_score': 84, '2020_score': 99 },
{ 'school_name': 'School B', '2015_score': 64, '2020_score': 89 },
{ 'school_name': 'School C', '2015_score': 87, '2020_score': 78 },
{ 'school_name': 'School D', '2015_score': 95, '2020_score': 98 },
])
df
| school_name | 2015_score | 2020_score | |
|---|---|---|---|
| 0 | School A | 84 | 99 |
| 1 | School B | 64 | 89 |
| 2 | School C | 87 | 78 |
| 3 | School D | 95 | 98 |
We could reshape (we'll do that down below), but we can learn a little more about ggplot by doing it the long way...
First, let's plot 2015's scores.
(
ggplot(df)
+ aes(y='2015_score', x=0, color='school_name')
+ geom_point()
)
<ggplot: (319714088)>
aes is smart and realizes that y='2015_score' refers to a column, while x=0 refers to a spot on the graph. While the y axis is related to 2015's score, x is always going to be at zero.
One thing you think might work is to just add in another set of points representing 2020's score. We'll do that by adding another aes and another geom_point() – we'll make these points be over at x=10 instead of at x=0.
(
ggplot(df)
+ aes(y='2015_score', x=0, color='school_name')
+ geom_point()
+ aes(y='2020_score', x=10, color='school_name')
+ geom_point()
)
<ggplot: (319763959)>
Unfortunately it only displays the second set of points! To add multiple points to the plotnine chart, we need to get very very specific, and give each geom_point its own aes. With that tiny change, our points show up in the right spots!
(
ggplot(df)
+ geom_point(aes(y='2015_score', x=0, color='school_name'))
+ geom_point(aes(y='2020_score', x=10, color='school_name'))
)
<ggplot: (319503357)>
To draw a line between them, we'll use geom_segment, which draws a line segment. This segment can take not only an x and a y, but also an optional xend and a yend – this makes it draw a point between (x, y) and (xend, yend).
(
ggplot(df)
+ geom_point(aes(y='2015_score', x=0, color='school_name'))
+ geom_point(aes(y='2020_score', x=10, color='school_name'))
+ geom_segment(aes(
x=0,
y='2015_score',
xend=10,
yend='2020_score',
color='school_name'
))
)
<ggplot: (319461949)>
Note that if you don't want color, you'll need to use group='school_name' instead of color='school_name', otherwise plotnine will try to connect all of the lines together.
(
ggplot(df)
+ geom_point(aes(y='2015_score', x=0))
+ geom_point(aes(y='2020_score', x=10))
+ geom_segment(aes(
x=0,
y='2015_score',
xend=10,
yend='2020_score',
group='school_name'
))
)
<ggplot: (319899859)>
The reshaped version¶
To perform the reshaped version, we'll just need to shift from wide to long (or "tidy" data) using .melt.
melted = df.melt(id_vars='school_name', value_name='score', var_name='year')
# Clean up the year column into just a number
melted.year = melted.year.str.replace("_score", "").astype(int)
melted.head()
| school_name | year | score | |
|---|---|---|---|
| 0 | School A | 2015 | 84 |
| 1 | School B | 2015 | 64 |
| 2 | School C | 2015 | 87 |
| 3 | School D | 2015 | 95 |
| 4 | School A | 2020 | 99 |
Once the data is in long format, now we can just say, "put the year on the x axis and the score on the y axis" and the chart gets built like magic! No more manual fudging with the details of ggplot.
(
ggplot(melted)
+ aes(x='year', y='score', group='school_name', color='school_name')
+ geom_point()
+ geom_line()
)
<ggplot: (319982577)>