Seaborn Cheat Sheet

01Setup & Themeimport · style · scale

import seaborn as sns★
The one import you always write.
import matplotlib.pyplot as plt
Still needed for show, savefig, fine-tuning.
import seaborn.objects as so
The newer declarative "objects" interface (card 12).
sns.set_theme()★
Apply seaborn's look — call once, up top.
sns.set_theme(style="whitegrid", context="talk", palette="deep")★
Look · sizing · colors in one call.
sns.set_style("darkgrid")
darkgrid·whitegrid·dark·white·ticks (default darkgrid).
sns.set_context("notebook")
Scale text/lines: paper·notebook·talk·poster.
sns.despine()
Remove the top & right spines for a cleaner frame.
tips = sns.load_dataset("tips")★
Built-in demo DataFrames: penguins, iris, titanic…

02The Data Modelroles, not coordinates

data=df★
Pass a tidy / long-form DataFrame; refer to columns by name.
x="total_bill", y="tip"★
Map columns to the axes by string name.
hue="sex"★
Color by a variable — auto legend. The workhorse channel.
size="size"
Scale point/line size by a variable.
style="smoker"
Vary marker shape / dash pattern by a variable.
col="time", row="day"★
Split into a grid of panels (facets) — figure-level only.
sns.scatterplot(data=df, x="a", y="b", hue="c")
The shape every seaborn function takes.

03Relationalrelplot · numeric vs numeric

sns.relplot(data=df, x=a, y=b)★fig-level
Default kind="scatter".
sns.relplot(…, kind="line")★
Line plot; auto-aggregates & draws a CI band.
sns.scatterplot(data=df, x=a, y=b)axes-level
Draws on the current Axes; returns it.
sns.lineplot(data=df, x=a, y=b)★
Averages repeated x and shades the CI.
sns.relplot(…, hue=c, size=n, style=c)
Layer several semantic channels at once.
sns.relplot(…, col=c, row=c)
Facet into a panel grid.
sns.lineplot(…, errorbar="sd")
CI band by default; use "sd", None, ("pi",95).

04Distributionsdisplot · shape of the data

sns.displot(data=df, x=a)★fig-level
Default kind="hist".
sns.histplot(data=df, x=a, bins=30)★
Histogram; binwidth= also works.
sns.histplot(…, kde=True)★
Overlay a smooth density curve on the bars.
sns.kdeplot(data=df, x=a, fill=True)★
Smooth density estimate (use fill, not shade).
sns.ecdfplot(data=df, x=a)
Empirical cumulative distribution — no binning choice.
sns.rugplot(data=df, x=a)
Little ticks for each raw observation.
sns.displot(…, multiple="stack")
With hue: layer·stack·dodge·fill.
sns.histplot(data=df, x=a, y=b)
Give x and y → a 2-D (bivariate) histogram.

05Categorical · pointscatplot · category vs number

sns.catplot(data=df, x=cat, y=num)★fig-level
Default kind="strip"; the gateway to all categorical kinds.
sns.stripplot(data=df, x=cat, y=num)
Jittered points, one strip per category.
sns.swarmplot(data=df, x=cat, y=num)
Non-overlapping "beeswarm" — best for small n.
sns.catplot(…, kind="swarm")
Same drawer, reached via the dispatcher.
sns.stripplot(…, hue=c, dodge=True)
Split each category by hue side-by-side.

06Categorical · spreadbox · violin · boxen

sns.boxplot(data=df, x=cat, y=num)★
Quartiles + whiskers + outliers.
sns.violinplot(data=df, x=cat, y=num)★
Box + a mirrored KDE density shape.
sns.violinplot(…, hue=c, split=True)
Half-violin per hue level — compact comparison.
sns.boxenplot(data=df, x=cat, y=num)
"Letter-value" box — shows tails better for large n.
sns.catplot(…, kind="box")
Any of these via kind= for free faceting.

07Categorical · estimatesbar · count · point

sns.barplot(data=df, x=cat, y=num)★
Bar of the mean (default) + CI — not a raw count.
sns.countplot(data=df, x=cat)★
Bar of row counts per category (a histogram of a category).
sns.pointplot(data=df, x=cat, y=num)
Point estimate + CI, joined — good for interactions.
sns.barplot(…, estimator="median", errorbar="ci")
Swap the summary statistic and the error bar.
sns.barplot(…, hue=c)
Grouped ("dodged") bars automatically.

08Regressionfit a trend line

sns.lmplot(data=df, x=a, y=b)★fig-level
Scatter + linear fit + CI, with faceting.
sns.regplot(data=df, x=a, y=b)axes-level
Same fit on one Axes; composes into subplots.
sns.lmplot(…, hue=c, col=c)
A separate fit per group / per panel.
sns.regplot(…, order=2)
Polynomial fit of the given degree.
sns.regplot(…, lowess=True)
Non-parametric locally-weighted smoother.
sns.regplot(…, logistic=True)
Logistic fit for a 0/1 outcome.
sns.residplot(data=df, x=a, y=b)
Plot residuals to check whether a fit is appropriate.

09Matrix plotsheatmap · clustermap

sns.heatmap(df.corr())★
Render a 2-D matrix as a grid of colored cells.
sns.heatmap(m, annot=True, fmt=".2f")★
Write each value inside its cell.
sns.heatmap(m, cmap="vlag", center=0)
Diverging colormap centered on zero (great for corr).
sns.clustermap(m)
Heatmap + hierarchical-clustering dendrograms, reordered.
sns.heatmap(m, linewidths=.5, cbar=False)
Cell borders; hide the color bar.
m = df.pivot(index=r, columns=c, values=v)
Reshape long → wide first; heatmap needs a matrix.

10Multi-plot gridsjoint · pair · FacetGrid

sns.pairplot(df, hue=c)★
Every pairwise scatter + per-variable diagonals.
sns.jointplot(data=df, x=a, y=b, kind="hex")★
Bivariate plot with marginal distributions.
sns.jointplot(…, kind="reg")
scatter·kde·hex·reg·hist.
g = sns.FacetGrid(df, col=c, row=c)
Build an empty panel grid to draw onto.
g.map_dataframe(sns.histplot, "x")
Draw the same plot into every panel.
sns.PairGrid(df).map_offdiag(sns.scatterplot)
Full control: different plot on diagonal vs off-diagonal.
g.add_legend(); g.set_axis_labels("x", "y")
Grid finishing touches.

11Color & palettesthe seaborn superpower

sns.color_palette("deep")★
Qualitative: deep·muted·pastel·bright·dark·colorblind.
sns.color_palette("husl", 8)
Evenly-spaced distinct hues for n categories.
sns.color_palette("crest", as_cmap=True)
Sequential (crest·flare·rocket·mako) for ordered data.
sns.color_palette("vlag")
Diverging (vlag·icefire) for data centered on a midpoint.
sns.scatterplot(…, hue=c, palette="Set2")
Any named palette, inline on the call.
sns.set_palette("pastel")
Set the default color cycle globally.
palette= without hue= → ignorednote
A palette needs a hue assignment to have anything to color.

12Objects interfaceimport seaborn.objects as so

so.Plot(df, x="a", y="b")★
The declarative base object — a grammar of graphics.
.add(so.Dot())★
A layer's Mark: Dot·Line·Bar·Area·Band·Text.
.add(so.Bar(), so.Agg())★
Mark + Stat: Agg·Est·Hist·KDE·Count·PolyFit.
.add(so.Dot(), so.Dodge(), so.Jitter())
Move: reposition — Dodge·Jitter·Stack·Shift.
.facet("c").pair(x=[…])
Small multiples & variable pairings, built in.
.scale(color="viridis")
Control how each data channel maps to a visual property.
.label(…).limit(…).theme(…)
Declarative finishing methods, chainable.
.layout(size=(7,5)).save("f.svg")
Size, then .save() / .show() / .on(ax) to embed.

13Styling the plotreach the matplotlib layer

ax = sns.boxplot(…)★
Axes-level returns a matplotlib Axes — use any of its methods.
ax.set(title="…", xlabel="…", ylim=(0,5))★
Batch matplotlib tweaks in one call.
g = sns.relplot(…)
Figure-level returns a grid g (FacetGrid) — style via g.
g.set_axis_labels("x", "y")
Seaborn's own label setter for the whole grid.
g.set_titles("{col_name}")
Template each panel's title from the facet value.
g.figure / g.axes
Reach the underlying matplotlib Figure / Axes array.
sns.relplot(…, height=4, aspect=1.5)★
Size figure-level plots with height×aspect, not figsize.

14Save & Showget it out of the notebook

plt.show()★
Render — automatic in notebooks, explicit in scripts.
plt.savefig("fig.png", dpi=300, bbox_inches="tight")★
Export an axes-level plot; tight trims whitespace.
g.savefig("fig.png", dpi=300)
Save a figure-level grid (call on g, not plt).
plot.save("fig.svg", bbox_inches="tight")
Objects-interface export.
# save BEFORE showorder
Some backends clear the figure on show(). Use .svg for vectors.

15Figure-level vs Axes-levelthe one big idea ★

scatterplot · histplot · boxplot · kdeplot · regplot★axes-level
Drop-in for matplotlib; draws on the current Axes via gca().
returns a matplotlib Axes · accepts ax=★
Compose it into your own plt.subplots() figure.
relplot · displot · catplot · lmplot★fig-level
One dispatcher per module; each builds its own figure.
returns a FacetGrid · ignores ax=★
Legend sits outside; col=/row= faceting for free.
kind="…" picks the axes-level drawer
displot(kind="kde") ≡ kdeplot().
explore → figure-level · compose → axes-level★
The practical rule of thumb.

16Gotchasthe usual traps

sns.relplot(…, ax=ax)no-op
Figure-level ignores ax= → you get an extra blank figure.
plt.subplots(); sns.displot(…)blank
A figure-level call after making axes leaves the axes empty.
boxplot(x=numeric)indexed
Categorical plots force a 0-indexed category axis — even for numbers.
palette= with no hue=
Does nothing — there's no variable to map colors to.
sns.set_theme()
Rewrites global matplotlib rcParams — affects all later plots.
figsize= on figure-level
Use height & aspect instead; figsize is ignored.
shade=Trueremoved
kdeplot(shade=) is gone — use fill=True.

seaborn cheat sheet

Four ideas worth a picture

figure-level vs axes-level

long-form vs wide-form data

one column → many visual channels

the objects grammar

Worth memorizing