Generate Publication-Quality Seaborn Visualizations

This skill applies Seaborn, the Python statistical visualization library built on matplotlib, to produce publication-quality statistical plots. It works from the core distinction between figure-level functions (displot, relplot, catplot - which create their own figure) and axes-level functions (histplot, scatterplot, boxplot - which plot onto existing axes):

import seaborn as sns
import matplotlib.pyplot as plt

### Figure-level: displot, relplot, catplot
g = sns.displot(data=df, x='value', hue='category', kind='hist', col='group')
g.fig.suptitle('Distribution Analysis')

### Axes-level: histplot, scatterplot, boxplot
fig, axes = plt.subplots(1, 2, figsize=(12, 5))
sns.histplot(data=df, x='value', hue='category', ax=axes[0])
sns.boxplot(data=df, x='category', y='value', ax=axes[1])

It covers statistical plot patterns for distribution analysis (histograms with KDE, violin plots with box overlays, ECDF plots, ridge plots via FacetGrid) and correlation/regression analysis (masked correlation heatmaps, regression plots with confidence intervals and annotated Pearson r/p values via scipy.stats). For customization it manages color palettes (husl, qualitative, diverging, sequential/cubehelix) and global context/style settings, and builds complex multi-panel layouts with FacetGrid across row and column facets. It also addresses performance for large datasets (sampling above 10,000 rows, rasterized scatter plots) and publication-ready output (theme and rcParams configuration, despine, high-DPI PDF/PNG export).

For troubleshooting and optimization it handles missing data explicitly - dropping nulls before plotting with dropna, or visualizing where nulls occur with a boolean heatmap - and manages memory on dense plots by tuning matplotlib's path chunksize and clearing figures with plt.clf() and plt.close('all') to prevent leaks across repeated plot generation.

Use this skill when you need statistically grounded, publication-quality plots - distribution comparisons, correlation heatmaps, regression plots with confidence intervals, or faceted multi-panel views - built with Seaborn's grammar-of-graphics approach rather than raw matplotlib. It is aimed at data storytelling and visual design best practices, not at building interactive or web-based dashboards.

Input is a pandas DataFrame (optionally reshaped with melt for long-format compatibility, or cast to categorical dtype with a defined category order). Output is a matplotlib figure or FacetGrid saved as a high-resolution PNG or PDF, styled with a colorblind-friendly palette, bold axis labels and titles, and despined for a clean publication look.

It is built directly on matplotlib (Seaborn's figures and axes are matplotlib objects, configured via plt.rcParams and pyplot calls) and pandas (categorical dtypes, melt for reshaping, DataFrame-native plotting). It uses scipy.stats.pearsonr to compute and annotate correlation statistics directly on regression plots.

Data scientists and analysts who need statistically sound, publication-ready charts - distribution comparisons, correlation and regression analysis, or multi-panel faceted views - straight from a pandas DataFrame.