seaborn

Use sns.boxplot(data=df, x='category', y='value') for quartiles and outliers, or sns.violinplot(data=df, x='category', y='value', inner='box') to combine a KDE with quartile information. Add hue='group' and split=True to compare two groups side by side.

Axes-level functions like scatterplot and boxplot draw on a single matplotlib Axes and accept an ax parameter. Figure-level functions like relplot, displot, and catplot manage an entire figure with built-in faceting via col and row, returning a FacetGrid object.

Seaborn is built on matplotlib and is best for statistical plots with automatic aggregation, confidence intervals, and attractive defaults. Drop to matplotlib when you need fine-grained control over axes, annotations, or custom layouts; the two integrate directly.

In seaborn 0.13, the ci parameter was replaced by errorbar in lineplot, barplot, and pointplot. Use errorbar=('ci', 95), 'sd', or 'se' instead. Regression functions like regplot and lmplot still accept ci.

Compute the matrix with df.corr(), then call sns.heatmap(corr, annot=True, fmt='.2f', cmap='coolwarm', center=0). Use numpy's triu to build a mask if you want to hide the redundant upper triangle.

Avoid seaborn for interactive visualizations, where plotly is a better fit, and for very large scatter datasets where swarmplot becomes computationally expensive. For millions of points, use hexbin aggregation or downsample first.