Themes and Context

Every chart you have drawn so far inherited a look from a single call: sns.set_theme(). That one line is Seaborn's global styling entry point. It does not attach to a chart or return an object — it changes matplotlib's defaults so that every plot you draw afterward picks up the same style, sizing, and colors. Set it once near the top of your work and forget about it.

This page is about steering that one call. set_theme takes four dials you will use constantly:

sns.set_theme(
    style="whitegrid",   # the canvas: background + gridlines
    context="notebook",  # the scale: how big fonts, lines, markers are
    palette="deep",      # the default colors (covered on its own page)
    font_scale=1.0,      # multiply all text sizes
)

We will spend this page on style (the canvas) and context (the scale), because those two carry most of the visual decision-making. We mention palette= only in passing — choosing colors well is a topic of its own, covered in depth on the color palettes page.

Style: the canvas behind your data

The style controls the non-data parts of the plot: the background color and whether gridlines are drawn. Seaborn ships five:

The choice is not cosmetic noise — it answers a real question: does my reader need to look up numbers, or just see the pattern? Gridded styles (darkgrid, whitegrid) lay down reference lines so a reader can trace a point back to the axis. Grid-free styles (white, ticks) remove that scaffolding and put all the visual weight on the data itself.

Let's draw the same simple plot under three styles so the difference is about the canvas, not the data. First whitegrid — the workhorse for most analysis:

Now white — the same points on a bare white canvas with no grid. Cleaner, more "finished", but you can no longer read an approximate body mass by eye:

And ticks — like white, but with short tick marks on the axes that give the eye just enough of a reference without a full grid. This is a favorite for publication figures:

A reasonable default rule: reach for whitegrid while you are analyzing (you often want to read values), and switch to white or ticks when you are presenting a clean figure where the shape is the message.

Context: scaling everything for the medium

A chart that looks perfectly balanced in a notebook can be unreadable when projected onto a screen at the front of a room — the text is too small, the lines too thin. The fix is context. It scales the sizes of elements (fonts, line widths, marker sizes) up or down as a group, without touching the style or the data.

There are four, from smallest to largest:

"paper"  <  "notebook" (the default)  <  "talk"  <  "poster"

The names tell you the intended medium. paper is sized for dense multi-panel figures in a publication; notebook for working at your screen; talk for slides; poster for something viewed from across a room.

Here is the default notebook context on a plain plot:

Now the same plot at talk context. The data, the axes, and the style are identical — only the element sizes grew. The labels, tick numbers, and even the points scale up so they stay legible from a distance:

Why bump it up for a slide? Because the figure will be shrunk to fit a slide and then blown back up by a projector, viewed from meters away. Text sized for arm's-length reading becomes a squint-fest on a wall. talk (or poster) pre-compensates by making everything bolder and bigger, so the audience reads it effortlessly.

A lighter frame with despine

Even after picking a style, plots carry a full rectangular border — a spine on all four sides. The top and right spines rarely add anything; they just box the data in. sns.despine() removes them, leaving only the left and bottom axes. The result feels lighter and more open, and it draws the eye toward the data rather than the frame.

despine() acts on the current figure, so call it after the plotting command:

The combination of style="ticks" + despine() is a classic publication look: a clean white canvas, small tick marks for reference, and only two axis lines. Notice that despine() does nothing on a gridded style where the spines are already faint — it shines on white and ticks.

Styling just one plot, temporarily

set_theme is global, which is exactly what you want most of the time — one consistent look for everything. But occasionally you want one plot to differ without disturbing the rest of your session. Seaborn gives you two context managers for that: axes_style (for the style) and plotting_context (for the context). Inside the with block the override applies; outside it, your global theme is untouched.

plotting_context works the same way for sizing — wrap a single plot in with sns.plotting_context("talk"): to scale up just that one figure. Reach for these only when you genuinely need a one-off; for the look of your whole analysis, set it once with set_theme and move on.

Your turn

Check your understanding

You can now control the overall look in one place: pick a style for the canvas, a context for the medium, and reach for despine() to lighten the frame. Next we go deep on the one dial we deliberately skipped — color palettes, and how to choose colors that inform rather than decorate.