Seaborn Basics — Statistical Charts Without the Boilerplate

Matplotlib gives you full control. Seaborn gives you statistical charts in two lines. The trade-off is worth it for EDA: seaborn handles aggregation, confidence intervals, grouping by color, and sensible defaults out of the box. The catch is understanding when you have a figure-level function and when you have an axes-level function — getting this wrong causes hours of confusion.

Learning Objectives

• Explain the difference between figure-level and axes-level seaborn functions
• Use hue to split data by a categorical variable
• Build distribution charts: histplot, kdeplot, boxplot, violinplot
• Build relational charts: scatterplot, lineplot
• Build categorical summary charts: barplot, countplot
• Combine seaborn charts with matplotlib customization

---

Setup

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

sns.set_theme(style="whitegrid", palette="muted")

Set the Theme Once, at the Top

sns.set_theme() sets global defaults for every chart in the session. Call it once at the top of your notebook. Common styles: "whitegrid" (clean with horizontal guides), "darkgrid", "ticks" (minimal), "white" (no grid).

---

Realistic Dataset

np.random.seed(42)
n = 300

df = pd.DataFrame({
    "study_hours": np.random.normal(4.5, 1.8, n).clip(0, 10),
    "sleep_hours": np.random.normal(7.0, 1.2, n).clip(4, 10),
    "score":       np.random.normal(68, 14, n).clip(20, 100),
    "attendance":  np.random.uniform(50, 100, n),
    "cohort":      np.random.choice(["Morning", "Evening", "Weekend"], n),
    "passed":      np.random.choice([True, False], n, p=[0.72, 0.28]),
})

# Make score positively correlated with study hours
df["score"] = (df["score"] + df["study_hours"] * 3.5).clip(20, 100)

---

The Most Important Concept: Figure-Level vs. Axes-Level

This is where almost everyone gets confused. Seaborn has two kinds of functions.

Axes-level functions (sns.scatterplot(), sns.histplot(), sns.boxplot(), etc.) draw onto a single ax. They work exactly like matplotlib — you can pass an ax= argument, combine them with other plots, and use all your matplotlib customization.

Figure-level functions (sns.relplot(), sns.displot(), sns.catplot()) create their own figure. They return a FacetGrid object, not a matplotlib axes. You cannot pass ax= to them. They are designed for faceting (splitting data into subplots by a column value).

Never Mix Figure-Level Functions With plt.subplots()

This is the most common seaborn mistake. Calling sns.relplot() inside a plt.subplots() block creates a separate figure and ignores your axes entirely. If you need to place a chart inside a subplot grid, use the axes-level equivalent: sns.scatterplot(ax=ax) instead of sns.relplot().

# WRONG — relplot ignores the ax you created
fig, ax = plt.subplots()
sns.relplot(data=df, x="study_hours", y="score")   # creates its OWN figure

# RIGHT — axes-level function respects your ax
fig, ax = plt.subplots(figsize=(7, 4))
sns.scatterplot(data=df, x="study_hours", y="score", ax=ax)
ax.set_title("Study Hours vs. Score")
plt.tight_layout()
plt.savefig("scatter_oo.png", dpi=150, bbox_inches="tight")  # or plt.show()

Quick reference:

Category	Figure-Level	Axes-Level Equivalent
Relational	relplot	scatterplot, lineplot
Distribution	displot	histplot, kdeplot, ecdfplot
Categorical	catplot	boxplot, violinplot, barplot, countplot, stripplot

---

Distribution Charts

histplot — Histogram with Optional KDE

fig, axes = plt.subplots(1, 2, figsize=(12, 4))

# Basic histogram
sns.histplot(data=df, x="score", bins=25, ax=axes[0],
             color="#0D9488", edgecolor="white")
axes[0].set_title("Score Distribution")
axes[0].set_xlabel("Exam Score")
axes[0].set_ylabel("Count")

# Histogram with KDE overlay, split by cohort
sns.histplot(data=df, x="score", hue="cohort", kde=True,
             bins=20, alpha=0.45, ax=axes[1])
axes[1].set_title("Score Distribution by Cohort")
axes[1].set_xlabel("Exam Score")
axes[1].set_ylabel("Count")

plt.tight_layout()
plt.savefig("histplot.png", dpi=150, bbox_inches="tight")  # or plt.show()

KDE — Kernel Density Estimate

kde=True overlays a smoothed curve representing the probability density. It makes the overall shape easier to read than a jagged histogram, especially when comparing multiple groups. The KDE is estimated from the data — it's not assuming a normal distribution.

kdeplot — Smooth Density Only

fig, ax = plt.subplots(figsize=(8, 4))

for cohort, color in zip(["Morning", "Evening", "Weekend"],
                          ["#0D9488", "#F59E0B", "#6366F1"]):
    subset = df[df["cohort"] == cohort]["score"]
    sns.kdeplot(subset, ax=ax, label=cohort, color=color,
                linewidth=2, fill=True, alpha=0.15)

ax.set_title("Score Distributions Are Similar Across Cohorts")
ax.set_xlabel("Exam Score")
ax.set_ylabel("Density")
ax.legend(title="Cohort")

plt.tight_layout()
plt.savefig("kdeplot.png", dpi=150, bbox_inches="tight")  # or plt.show()

boxplot — Spread and Outliers

fig, ax = plt.subplots(figsize=(8, 4))

sns.boxplot(data=df, x="cohort", y="score", ax=ax,
            palette="muted", order=["Morning", "Evening", "Weekend"])

ax.set_title("Score Distribution by Cohort")
ax.set_xlabel("Cohort")
ax.set_ylabel("Exam Score")

plt.tight_layout()
plt.savefig("boxplot.png", dpi=150, bbox_inches="tight")  # or plt.show()

A boxplot shows: median (center line), interquartile range (box), whiskers (1.5 × IQR), and individual outliers as dots.

violinplot — Distribution Shape + Box

A violin plot combines a KDE with a boxplot. It shows the full distribution shape, not just the summary statistics. Use it when the shape matters — a bimodal distribution looks like a normal distribution in a boxplot.

fig, axes = plt.subplots(1, 2, figsize=(12, 4))

sns.boxplot(data=df, x="cohort", y="score", ax=axes[0], palette="muted")
axes[0].set_title("Boxplot (hides shape)")
axes[0].set_xlabel("Cohort")
axes[0].set_ylabel("Score")

sns.violinplot(data=df, x="cohort", y="score", ax=axes[1],
               palette="muted", inner="box", cut=0)
axes[1].set_title("Violin (shows shape)")
axes[1].set_xlabel("Cohort")
axes[1].set_ylabel("Score")

plt.tight_layout()
plt.savefig("violin_vs_box.png", dpi=150, bbox_inches="tight")  # or plt.show()

---

Relational Charts

scatterplot — Relationship Between Two Numeric Variables

fig, ax = plt.subplots(figsize=(8, 5))

sns.scatterplot(data=df, x="study_hours", y="score",
                hue="cohort", style="passed",
                alpha=0.65, s=50, ax=ax)

ax.set_title("More Study Hours → Higher Scores")
ax.set_xlabel("Study Hours per Day")
ax.set_ylabel("Exam Score")
ax.legend(title="Cohort / Passed", bbox_to_anchor=(1.02, 1), loc="upper left")

plt.tight_layout()
plt.savefig("scatterplot.png", dpi=150, bbox_inches="tight")  # or plt.show()

Overplotting in Scatter Plots

When you have thousands of points, scatter plots become solid blobs. Solutions: set alpha=0.2 (transparency), use sns.kdeplot(kind="hex") for 2D density, or bin and aggregate before plotting. A scatter plot with 10,000 fully opaque dots tells you nothing.

lineplot — Trend Over Time with Confidence Interval

sns.lineplot automatically computes the mean and 95% confidence interval when you have multiple observations per x value.

# Simulate daily signups with multiple readings per day
np.random.seed(0)
days = np.repeat(np.arange(1, 31), 5)
signups = (days * 1.8 + np.random.normal(0, 4, len(days))).clip(0)

time_df = pd.DataFrame({"day": days, "signups": signups,
                         "channel": np.random.choice(["Organic", "Paid"], len(days))})

fig, ax = plt.subplots(figsize=(10, 4))

sns.lineplot(data=time_df, x="day", y="signups",
             hue="channel", ax=ax, linewidth=2)

ax.set_title("Daily Signups by Acquisition Channel (shaded = 95% CI)")
ax.set_xlabel("Day of Month")
ax.set_ylabel("Signups")
ax.legend(title="Channel")

plt.tight_layout()
plt.savefig("lineplot_ci.png", dpi=150, bbox_inches="tight")  # or plt.show()

---

Categorical Summary Charts

barplot — Aggregated Mean with Confidence Interval

sns.barplot computes the mean by default and draws error bars representing the 95% CI. It is not a count chart — it's an estimation chart.

fig, ax = plt.subplots(figsize=(8, 4))

sns.barplot(data=df, x="cohort", y="score",
            hue="passed", ax=ax,
            order=["Morning", "Evening", "Weekend"])

ax.set_title("Average Score by Cohort and Pass Status")
ax.set_xlabel("Cohort")
ax.set_ylabel("Average Score")
ax.legend(title="Passed")

plt.tight_layout()
plt.savefig("barplot.png", dpi=150, bbox_inches="tight")  # or plt.show()

countplot — Count of Observations Per Category

fig, ax = plt.subplots(figsize=(7, 4))

sns.countplot(data=df, x="cohort", hue="passed",
              order=["Morning", "Evening", "Weekend"], ax=ax)

ax.set_title("Student Count by Cohort and Pass Status")
ax.set_xlabel("Cohort")
ax.set_ylabel("Number of Students")
ax.legend(title="Passed")

plt.tight_layout()
plt.savefig("countplot.png", dpi=150, bbox_inches="tight")  # or plt.show()

---

Figure-Level Functions — Faceting

Use sns.relplot(), sns.displot(), or sns.catplot() when you want to split data into separate subplots by a column. Do not use them when you just want a single chart.

g = sns.relplot(
    data=df,
    x="study_hours", y="score",
    col="cohort",
    hue="passed",
    alpha=0.6, s=40,
    height=4, aspect=1.1
)

g.set_titles("Cohort: {col_name}")
g.set_axis_labels("Study Hours", "Score")
g.figure.suptitle("Score vs. Study Hours by Cohort", y=1.03)

g.figure.savefig("facet_scatter.png", dpi=150, bbox_inches="tight")  # or plt.show()

---

The hue Parameter

hue is seaborn's most powerful parameter. Pass it a categorical column and seaborn automatically: - assigns a different color to each category - adds a legend - keeps all other chart properties consistent

fig, axes = plt.subplots(1, 3, figsize=(14, 4))

# hue on scatter
sns.scatterplot(data=df, x="attendance", y="score",
                hue="cohort", alpha=0.5, ax=axes[0])
axes[0].set_title("Scatter with hue")

# hue on histogram
sns.histplot(data=df, x="score", hue="cohort",
             bins=20, alpha=0.5, ax=axes[1])
axes[1].set_title("Histogram with hue")

# hue on box
sns.boxplot(data=df, x="cohort", y="score",
            hue="passed", ax=axes[2])
axes[2].set_title("Boxplot with hue")

plt.tight_layout()
plt.savefig("hue_examples.png", dpi=150, bbox_inches="tight")  # or plt.show()

---

Combining Seaborn and Matplotlib

Seaborn creates the chart. Matplotlib adjusts the details. This is the standard workflow.

fig, ax = plt.subplots(figsize=(9, 5))

sns.scatterplot(data=df, x="study_hours", y="score",
                hue="cohort", alpha=0.5, s=50, ax=ax)

# Add a regression line manually
m, b = np.polyfit(df["study_hours"], df["score"], 1)
x_line = np.linspace(df["study_hours"].min(), df["study_hours"].max(), 100)
ax.plot(x_line, m * x_line + b, color="black", linewidth=2,
        linestyle="--", label=f"Trend: y={m:.1f}x+{b:.1f}")

# Matplotlib customization
ax.set_title("Study Hours Predict Scores Across All Cohorts", fontsize=13)
ax.set_xlabel("Study Hours per Day", fontsize=11)
ax.set_ylabel("Exam Score", fontsize=11)
ax.legend(title="Cohort", framealpha=0.9)
ax.annotate("Each point = one student", xy=(0.02, 0.95),
            xycoords="axes fraction", fontsize=9, color="gray")

plt.tight_layout()
plt.savefig("seaborn_matplotlib_combo.png", dpi=150, bbox_inches="tight")  # or plt.show()

---

Practice Exercises

Warm-up: Using the df dataset, create a histplot of study_hours with kde=True. Split by cohort using hue. Give it a title and axis labels.

Main: Create a 1×2 subplot. Left: a violinplot of score by cohort. Right: a barplot of mean score by cohort split by passed using hue. Use tight_layout().

Stretch: Create a pairplot of the numeric columns study_hours, sleep_hours, score, and attendance, colored by cohort. Observe which pairs show the strongest visual correlation. Write a one-sentence interpretation for each.

---

Interview Questions

Q: What is the difference between figure-level and axes-level seaborn functions?

Show answer

Axes-level functions (e.g., scatterplot, histplot, boxplot) draw onto a single matplotlib Axes object. You can pass ax=ax to place them in a subplot grid and use matplotlib to customize them. Figure-level functions (e.g., relplot, displot, catplot) create their own figure and return a FacetGrid. They are designed for faceting across subplots by a column value. You cannot pass ax= to a figure-level function.

Q: What does hue do in seaborn?

Show answer

hue maps a categorical variable to color. Seaborn assigns a distinct color to each category, plots each group separately, and automatically generates a legend. It is the simplest way to add a third dimension to a two-axis chart.

Q: When would you choose a violin plot over a box plot?

Show answer

When the shape of the distribution matters. A box plot summarizes five statistics (min, Q1, median, Q3, max). A bimodal distribution — two separate clusters of values — looks identical to a normal distribution in a box plot. A violin plot shows the full KDE of the distribution, so you can see bimodality, skewness, and gaps that a box plot hides.

Q: sns.barplot() shows confidence interval error bars by default. What does this mean?

Show answer

sns.barplot() computes the mean of the y variable for each x category, then uses bootstrapping to estimate the 95% confidence interval around each mean. The error bar shows this interval — a taller error bar means more uncertainty (higher variance or fewer data points). It is different from a simple bar chart, which shows a raw value with no uncertainty.

---

Key Takeaways

• Axes-level functions (scatterplot, histplot, boxplot, etc.) work inside plt.subplots() grids. Figure-level functions (relplot, displot, catplot) create their own figure — never mix them with plt.subplots().
• hue is seaborn's most powerful feature for exploratory analysis — use it to reveal group-level patterns.
• violinplot shows distribution shape. boxplot shows summary statistics. Use violin when shape matters.
• Seaborn creates the chart. Matplotlib customizes it. Combine both freely.

---

Previous: Line, Bar, and Histogram Charts | Next: Heatmaps