visualization_lecture

Seaborn

Learning objectives

• Understand Seaborn’s design principles and how it differs from matplotlib.
• Know the major plot types and the common Seaborn functions for each.
• Distinguish figure-level and axes-level APIs (e.g., displot vs histplot).
• Be able to create informative plots with built-in datasets and style controls.
• Learn basic composition: FacetGrid, PairGrid, combining Seaborn + Matplotlib.
• Gain three hands-on mini-exercises to apply learned techniques.

---

Quick outline (timed)

1. Introduction & design philosophy — 5 min
2. Seaborn basics: data formats, theme/style — 10 min

3. Plot categories and representative functions — 25 min

   • Relational (scatter/line)
   • Categorical
   • Distribution
   • Regression & statistical
   • Matrix / multivariate
4. Grids & figure-level plots (FacetGrid, PairGrid) — 8 min
5. Practical tips, pitfalls, combining with Matplotlib — 7 min
6. Exercises & resources — 5 min

---

1. Introduction & design principles (5 min)

• Seaborn is a high-level API built on top of matplotlib, designed for statistical data visualization.
• Works best with tidy (long-form) data (one observation per row, columns are variables).
• Offers default themes, color palettes, and convenient statistical aggregations.

• Two main API styles:

  • Figure-level functions (create entire figure + facets) — displot, relplot, catplot, pairplot, lmplot.
  • Axes-level functions (plot on an existing Axes) — scatterplot, histplot, boxplot, violinplot, heatmap, etc.

---

2. Seaborn basics: data formats, theme, imports (10 min)

import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd

sns.__version__  # ensure recent version (0.11+ has many features)

# load example datasets
tips = sns.load_dataset("tips")
iris = sns.load_dataset("iris")
penguins = sns.load_dataset("penguins")

Data forms

• long (tidy): each row = observation. e.g. tips (total_bill, tip, sex, smoker, day, time)
• wide: multiple related measurement columns (e.g., columns for each time point). Some seaborn functions accept wide-form directly.

Themes & context

sns.set_theme(style="whitegrid", context="notebook")
# style: whitegrid, darkgrid, white, ticks, dark
# context: paper, notebook, talk, poster

# control rc parameters
sns.set(rc={"figure.figsize": (8, 5)})

Color palettes

sns.color_palette("deep")
sns.color_palette("crest", as_cmap=True)

# categorical vs sequential vs diverging
sns.palplot(sns.color_palette("tab10"))

---

3. Plot categories & representative functions (25 min)

We’ll list functions by purpose with short code examples.

A. Relational plots — show relationships between variables

Axes-level: scatterplot, lineplot.

sns.scatterplot(data=tips, x="total_bill", y="tip", hue="time", style="smoker")

sns.lineplot(data=some_time_series_df, x="date", y="value", hue="category")

Figure-level: relplot(kind="scatter" / "line") — convenience wrapper that returns FacetGrid.

sns.relplot(data=tips, x="total_bill", y="tip", col="day", hue="sex", kind="scatter")

Notes: scatterplot accepts size, hue, style for multidimensional encoding. Use alpha and s for density/marker size.

B. Categorical plots — comparing categories

• stripplot, swarmplot (jittered points)
• boxplot, violinplot (distribution summaries)
• barplot (aggregate with CI), countplot (counts)
• pointplot (point estimates with CI)

sns.boxplot(data=tips, x="day", y="total_bill")
sns.violinplot(data=tips, x="day", y="total_bill", hue="sex", split=True)

sns.barplot(data=tips, x="day", y="total_bill", estimator=sum)
sns.countplot(data=tips, x="day")

Figure-level: catplot(kind=...) — good for faceting.

sns.catplot(data=tips, x="day", y="tip", kind="violin", hue="sex", col="time")

When to use: use boxplot for robust summaries, violinplot for showing full distribution, strip/swarm for raw points.

C. Distribution plots — one or two variables

• histplot, kdeplot, ecdfplot, displot (figure-level combining hist/kde)

sns.histplot(data=iris, x="sepal_length", kde=True)
sns.kdeplot(data=iris, x="sepal_length", hue="species", fill=True)

sns.displot(data=iris, x="sepal_length", col="species", kde=True)

Notes: displot returns a FacetGrid so you can facet easily.

D. Regression & statistical models

• regplot (axes-level), lmplot (figure-level + faceting)

sns.regplot(data=tips, x="total_bill", y="tip", scatter_kws={"s": 10})

sns.lmplot(data=tips, x="total_bill", y="tip", hue="smoker", col="day")

Options: lowess smoothing, polynomial order, ci for confidence intervals.

E. Multivariate / matrix plots

• heatmap (matrices/correlations)
• clustermap (hierarchical clustering heatmap)

corr = iris.corr()
sns.heatmap(corr, annot=True, fmt='.2f', square=True)

sns.clustermap(iris.select_dtypes('number'), metric='euclidean', method='average')

Notes: Good for showing variable correlation structures or distance matrices.

F. Pairwise relationships

• pairplot (quick pairs), PairGrid (customizable)
• jointplot, JointGrid for bivariate + marginal distributions

sns.pairplot(iris, hue="species", corner=True)

sns.jointplot(data=iris, x="sepal_length", y="sepal_width", kind="hex")

When to use: exploratory data analysis to inspect pairwise relationships and marginal distributions.

---

4. Grids & figure-level plots: Faceting (8 min)

• FacetGrid is the core: map plotting functions across facets (rows, cols, hue).

g = sns.FacetGrid(tips, col="time", row="smoker", margin_titles=True)
g.map(sns.scatterplot, "total_bill", "tip")

• relplot, catplot, displot, lmplot, pairplot are built-in figure-level functions that create and use a FacetGrid internally.

Tips: Use sharex/sharey to control axis sharing; use col_wrap to wrap many facets into rows.

---

5. Practical tips & common pitfalls (7 min)

Figure-level vs Axes-level

• If you need a single Axes in an existing figure (e.g., subplot layout with plt.subplots()), use axes-level functions and pass ax= or operate on ax.
• Use figure-level when you want seaborn to control the figure and produce faceted plots.

Long vs Wide data

• Prefer tidy (long) data. If your data is wide, use pd.melt() to reshape.

Combining with Matplotlib

fig, ax = plt.subplots()
sns.boxplot(data=tips, x="day", y="total_bill", ax=ax)
ax.set_title("Total bill by day")

Controlling aesthetics

• sns.set_theme() is recommended.
• Use palette argument for color control.
• Use legend/ax.legend() to relocate or format legends.

Performance

• For large datasets, use alpha, s, or hex/bin plots: sns.histplot(..., bins=100, pthresh=0.05, cmap='mako') or kind='hex' in jointplot.

Statistical interpretation

• Know what estimator and ci mean in barplot/pointplot — defaults are mean and bootstrapped 95% CI.

---

6. Mini exercises (5 min explanation, student time can follow outside this lecture)

Exercise 1 (Exploration, 15–30 min)

• Use penguins or tips data.
• Produce a pairplot colored by species/sex.
• Identify interesting pairwise correlations and summarise them in 5 bullets.

Exercise 2 (Figure composition)

• Create a figure with 2x2 subplots. Top-left: heatmap of correlations for numeric variables. Top-right: boxplot by category. Bottom row: scatterplot with regression line.
• Export as PNG and add titles and axis labels.

Exercise 3 (Communication)

• Given tips, make a single publication-quality figure that shows how tip varies by day and time, and includes raw points (jitter), a violin or box for distribution, and a summary stat (mean) as points.

---

7. Useful functions & quick reference (cheat sheet)

Relational: scatterplot, lineplot, relplot

Categorical: catplot, boxplot, violinplot, barplot, countplot, stripplot, swarmplot, pointplot

Distribution: histplot, kdeplot, ecdfplot, displot

Regression: regplot, lmplot

Pairwise / Joint: pairplot, PairGrid, jointplot, JointGrid

Matrix: heatmap, clustermap

Utilities: load_dataset, set_theme, color_palette, palplot, despine

---

8. Further reading & resources (links you can include in repo)

• Official documentation & gallery (seaborn.pydata.org) — gallery is excellent for examples.
• Seaborn API reference pages (look up each function above).
• Practical tutorials: Jake VanderPlas’ chapters on visualization; statistical visualization sections in Python data science books.

---

9. Appendix: Example code snippets (copy/paste friendly)

# Quick exploratory EDA
import seaborn as sns
sns.set_theme(style='whitegrid')

# 1. Pairplot
sns.pairplot(sns.load_dataset('iris'), hue='species')

# 2. Faceted scatter
sns.relplot(data=sns.load_dataset('tips'), x='total_bill', y='tip', col='day', hue='sex')

# 3. Combined categorical plot
ax = sns.violinplot(data=tips, x='day', y='total_bill', inner=None)
sns.stripplot(data=tips, x='day', y='total_bill', color='k', size=3, jitter=True, ax=ax)

# 4. Heatmap of correlations
import numpy as np
corr = sns.load_dataset('penguins').select_dtypes('number').corr()
sns.heatmap(corr, annot=True, cmap='vlag')

Next: Colour theory

This site is open source. Improve this page.