Scatter Plots and Pair Plots

Distribution plots show one variable at a time. When you want to understand the relationship between two numeric variables, you need scatter plots and related visualisations. These help you detect correlation (do both variables increase together?), clusters (are there subgroups in the data?), and outliers (are there unusual points far from the main cloud?). Seaborn makes all of these easy with scatterplot and pairplot.

sns.scatterplot(data, x, y) plots each observation as a point at its (x, y) coordinates. Unlike Matplotlib's plt.scatter, Seaborn's version automatically links to a DataFrame and integrates with hue, size, and style semantics. The result is informative multi-variable plots with a single function call and an automatic legend.

The hue parameter assigns a colour to each point based on a third variable — either categorical or numeric. When hue is a categorical column (like 'smoker'), Seaborn uses distinct colours. When hue is a numeric column, it uses a sequential colour map. This lets you see three variables at once on a 2D plot without adding a third axis.

Beyond hue, Seaborn scatter plots support size (point area encodes a numeric variable) and style (marker shape encodes a categorical variable). Using all three semantics simultaneously can reveal complex multi-variable patterns but risks overloading the viewer. A practical guideline is to use hue first (most salient), style second, and size only when necessary.

sns.regplot() draws a scatter plot and overlays a linear regression line with a shaded 95% confidence band. This is useful for visualising the strength and direction of the linear relationship between two variables. The confidence band narrows in regions with dense data and widens at the edges where fewer points inform the fit.

sns.lmplot() is the figure-level version of regplot. It supports hue (separate regression lines per group in the same panel) and col/row (separate panels). This is powerful for testing whether the relationship between two variables differs across subgroups — for example, whether the bill-to-tip relationship is steeper for dinner than for lunch.

sns.pairplot(df) creates a grid of scatter plots for every pair of numeric columns in the DataFrame, with distribution plots on the diagonal. This is the fastest way to visualise all pairwise relationships in a dataset in one call. For a DataFrame with n numeric columns, pairplot creates an n×n grid. It is most practical when n is between 3 and 8 — larger grids become too small to read.

The diag_kind parameter controls what appears on the diagonal of the pairplot grid. Use 'hist' for histograms or 'kde' for kernel density estimates. The diagonal shows the univariate distribution of each variable, while the off-diagonal panels show pairwise scatter plots. You can also use corner=True to show only the lower triangle, halving the number of panels and reducing redundancy.

sns.PairGrid gives you full control over which plot type appears in each section of the matrix. Use g.map_upper(), g.map_lower(), and g.map_diag() to assign different functions (like regplot on the upper triangle and kdeplot on the lower). This produces publication-quality plots where each panel conveys distinct information about the variable pair.

When reading a scatter plot, estimate the direction (does the cloud slope up or down?), strength (how tightly clustered are the points around the trend?), and form (linear or curved?) of the relationship. Correlation coefficient r ranges from -1 (perfect negative) to +1 (perfect positive), with 0 meaning no linear relationship. Remember: correlation is not causation, and a non-linear relationship can have r ≈ 0.

sns.jointplot() combines a central bivariate plot with marginal univariate plots on the top and right edges. Pass kind='scatter', 'hex', 'kde', or 'reg'. The hex kind is useful when you have thousands of overlapping points — it bins them into hexagons and uses colour to show density, solving the overplotting problem that makes scatter plots unreadable for large datasets.

Test your understanding of Seaborn scatter and pair plots from this lesson.

In this lesson you learned: sns.scatterplot encodes up to four variables using x, y, hue, size, and style, sns.regplot/lmplot overlays a regression line to quantify linear relationships, and sns.pairplot produces an all-pairs grid ideal for exploratory analysis of multi-variable datasets. Next up we explore heatmaps for visualising correlation matrices.