Generate Statistical Visualizations with Seaborn

Seaborn Statistical Visualization covers Python's seaborn library for publication-quality statistical graphics, built on five design principles - dataset-oriented plotting directly from DataFrames, automatic semantic mapping of data to visual properties, built-in statistical aggregation and confidence intervals, publication-ready aesthetic defaults, and full matplotlib compatibility. It spans both of seaborn's plotting interfaces: the traditional function interface (axes-level functions like scatterplot, histplot, and boxplot for single-axes plots, and figure-level functions like relplot, displot, and catplot that manage an entire faceted figure) and the modern, ggplot2-like seaborn.objects declarative interface for composable, layered visualizations.

Use it for publication-quality statistical graphics straight from tabular data, exploring multivariate relationships or distributions with minimal code, or when seaborn's dataset-oriented API and statistical defaults on top of matplotlib fit the job better than raw matplotlib. It covers five plot families by name - relational (scatterplot, lineplot, relplot), distribution (histplot, kdeplot, ecdfplot, rugplot, displot, jointplot, pairplot), categorical (stripplot, swarmplot, boxplot, violinplot, boxenplot, barplot, pointplot, countplot, catplot), regression (regplot, lmplot, residplot), and matrix (heatmap, clustermap) - plus three grid objects (FacetGrid, PairGrid, JointGrid) for custom multi-panel layouts. Choosing the right one depends on data shape: continuous-x/continuous-y favors scatterplot, lineplot, or regplot; one continuous variable favors histplot, kdeplot, or ecdfplot; and correlation matrices favor heatmap or clustermap.

Input: long-form data ("tidy," one row per observation, one column per variable, preferred and required for full flexibility) or wide-form data (columns spread across variables, convertible to long-form via df.melt()), plus semantic mappings (hue, size, style) and, for figure-level functions, col/row faceting parameters. Output: a matplotlib figure or axes object, styled via set_theme()'s style options (darkgrid, whitegrid, dark, white, ticks) and context scales (paper, notebook, talk, poster), colored via qualitative (deep, muted, pastel, colorblind), sequential (rocket, mako, viridis), or diverging (vlag, icefire, coolwarm) palettes.

sns.set_theme(style='ticks', context='paper', font_scale=1.1)

g = sns.catplot(data=df, x='treatment', y='response',
                col='cell_line', kind='box', height=3, aspect=1.2)
g.set_axis_labels('Treatment Condition', 'Response (μM)')
g.set_titles('{col_name}')
sns.despine(trim=True)

g.savefig('figure.pdf', dpi=300, bbox_inches='tight')

It also documents seven best practices - named columns over unnamed arrays, matching plot type to data shape, figure-level functions for faceting instead of manual subplots, semantic mappings for extra dimensions, explicit control over statistical estimators like mean vs. median and CI vs. standard deviation, matplotlib fine-tuning via the returned Axes or Figure object, and high-DPI or vector export for publication - and five named troubleshooting fixes: repositioning a legend placed outside the figure by default, rotating overlapping tick labels, resizing a too-small figure via height/aspect or figsize, choosing a more distinct color palette, and adjusting bw_adjust when a KDE is too smooth or too jagged.

Built directly on matplotlib - every seaborn plot returns a matplotlib Axes, Figure, or grid object that accepts further matplotlib customization - and pandas DataFrames as its native data structure; ships three bundled reference files, function_reference.md, objects_interface.md, and examples.md, for deeper function signatures and use-case patterns.

Data scientists and analysts who want statistically-aware, publication-ready plots - multivariate scatter, line, distribution, categorical, regression, or matrix charts, faceted small multiples, or custom multi-panel grids - generated with far less code than raw matplotlib, while still able to drop down to matplotlib for fine control when needed.