Programming courses often start with control structures and data structures. Education research has shown the value of starting with visualization instead, because it gives students early results that are compelling and often fun. We will keep refining our visualization skills throughout the course, and much of that improvement comes through repetition.
The Grammar of Graphics¶
We use plotnine, which implements the
Grammar of Graphics in Python. The grammar
provides a powerful abstraction for data visualization. The theory is decades old, and
it became mainstream in data science through Hadley Wickham’s ggplot2 package for R.
Python libraries adopted the idea more recently, and it now appears in plotnine,
altair and the seaborn object interface.
Mapping data to aesthetics¶
The central principle of the grammar of graphics is that a plot expresses mappings from data to aesthetics such as color, size and x and y position.
import ibis
from plotnine import ggplot, aes, geom_line, labs
con = ibis.duckdb.connect()
co2 = con.read_csv("data/co2.csv")
co2.head().to_pandas()This dataset is in long form. It has a decimal_date column, a name column that
identifies which series each row belongs to, and a value column. We can map the
categorical values in name to color:
(
ggplot(co2.to_pandas(), aes(x="decimal_date", y="value", color="name"))
+ geom_line()
+ labs(x="Year", y="CO2 (ppm)")
)The plot has three parts: the data and its mapping of columns to aesthetics (aes), the
geometry that draws them (geom_line), and the labels. We combine the layers with +.
Manual styling¶
Older plotting engines such as matplotlib, and the original seaborn function
interface, do not follow these principles. The equivalent plot in that style looks like
this:
# the older style, which we avoid
import seaborn as sns
wide = co2.pivot_wider(names_from="name", values_from="value").to_pandas()
sns.lineplot(wide, x="decimal_date", y="average", color="blue")
sns.lineplot(wide, x="decimal_date", y="deseasonalized", color="red")Here x and y still refer to columns, but each color is set by hand. The code does not
record any relationship between color and the structure of the data, so changing the
color scheme means editing every call. This approach is harder to generalize and more
prone to error.
The manual version is slightly shorter for a single line, but it grows quickly as we add series. A third series needs a third, nearly identical call. In the grammar version, a third series needs no new code because it is already in the data.
Tidy data¶
The grammar version works because the data are tidy: each column is a variable and each row is an observation. The identity of each series is stored in a column, so the grammar can map it to an aesthetic.
This layout is sometimes called long form, and more formally Codd’s third normal form. It
takes some getting used to, and it often requires a transformation, because many datasets
in the wild do not follow it. Older and smaller datasets are especially likely to use a
wide layout. pivot_longer() and pivot_wider() convert between the two shapes.
The grammar and tidy data reinforce each other. A well-designed plotting abstraction encourages a well-organized data structure, and a well-organized data structure turns the plot into a single line of code. A good abstraction makes the correct approach easy and the incorrect approach awkward.
Default palettes¶
The grammar version also uses a default color palette informed by visualization research.
In the manual version, "blue" and "red" were chosen without any evidence. Good defaults
are a real feature of a package, and when we override them we should do so on purpose.
Reference¶
plotnine gallery. Find a chart that looks like the one you want, then read its code.
The Grammar of Graphics, Wilkinson