Plotnine Grammar of Graphics

  • ID: VISPY-005
  • Type: Lesson
  • Audience: Public
  • Theme: Grammar-of-graphics thinking

Matplotlib and Seaborn are excellent for building figures directly.

Plotnine is useful when you want to think in layers:

The point of this lesson is not to adopt a new library. It is to learn a mental model that improves plotting in any library.

Setup

import pandas as pd

from plotnine import (
    ggplot, aes, geom_point, geom_smooth,
    geom_boxplot, facet_wrap, labs,
    theme_light, theme, element_text
)

df = pd.read_csv("data/cdi-student-outcomes.csv")
print(df.head())
     group  test_prep  study_hours  math_score  reading_score  writing_score
0  Group B  completed          3.9          58             64             51
1  Group A       none          7.7          67             85             61
2  Group A       none          9.3          83             65             73
3  Group A       none          3.9          60             67             48
4  Group A       none          8.3          68             63             47

Core idea: mappings vs style

In grammar-of-graphics, you separate:

  • what the data encodes (mappings)
  • what is only styling (colors, sizes, labels)

Example:

  • mapping: x=study_hours, y=math_score, color=test_prep
  • styling: point transparency, line thickness

Relationship plot (layered)

p = (
    ggplot(df, aes(x="study_hours", y="math_score", color="test_prep"))
    + geom_point(alpha=0.55)
    + geom_smooth(method="lm", se=False)
    + labs(
        title="Study Hours vs Math Score",
        subtitle="Trend shown separately by test preparation",
        x="Study hours per week",
        y="Math score",
        color="Test prep"
    )
    + theme_light(base_size=14)
    + theme(
        plot_title_position="plot",
        plot_title=element_text(ha="center", weight="bold", margin={"b": 6}),
        plot_subtitle=element_text(ha="center", margin={"b": 10}),
        legend_position="top",
    )
)

p

Group comparison (summary)

p = (
    ggplot(df, aes(x="test_prep", y="math_score"))
    + geom_boxplot()
    + labs(
        title="Math Score by Test Preparation",
        subtitle="Boxplots summarize differences across groups",
        x="Test preparation",
        y="Math score"
    )
    + theme_light(base_size=14)
    + theme(
        plot_title_position="plot",
        plot_title=element_text(ha="center", weight="bold", margin={"b": 6}),
        plot_subtitle=element_text(ha="center", margin={"b": 10}),
    )
)

p

Faceting: small multiples

p = (
    ggplot(df, aes(x="study_hours", y="math_score"))
    + geom_point(alpha=0.55)
    + geom_smooth(method="lm", se=False)
    + facet_wrap("~group")
    + labs(
        title="Study Hours vs Math Score by Group",
        subtitle="Same plot repeated by group for disciplined comparison",
        x="Study hours per week",
        y="Math score"
    )
    + theme_light(base_size=14)
    + theme(
        plot_title_position="plot",
        plot_title=element_text(ha="center", weight="bold", margin={"b": 8}),
        plot_subtitle=element_text(ha="center", margin={"b": 12}),
    )
)

p

Exporting a Plotnine figure

from pathlib import Path
Path("figures").mkdir(exist_ok=True)

out = Path("figures") / "05_plotnine_facet.png"
p.save(out, dpi=160, verbose=False)
str(out)
'figures/05_plotnine_facet.png'

Key Takeaways

  • Grammar-of-graphics is a mental model, not a brand.
  • Separate mappings (meaning) from styling (appearance).
  • Layering helps you modify plots without restarting.
  • Faceting is one of the cleanest ways to compare groups.