Relationships: scatterplot with hue, regplot, heatmap
Many charts at once: pairplot and FacetGrid
🏢 In the real world
A school chain wants to know: “Does tuition really help? Do students with internet at home score more? Is study time or attendance more important?” Seaborn answers each question with one line of code – and shows the uncertainty too.
Why Seaborn?
Matplotlib
Full control, more code
You group and aggregate yourself
Best for custom layouts
Seaborn
Works directly with DataFrames
hue= splits by category automatically
Statistics (mean, CI, KDE) built in
Distributions
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
st = pd.read_csv("students_performance.csv")
fig, ax = plt.subplots(1, 2, figsize=(11, 4))
sns.histplot(data=st, x="average", hue="internet", kde=True, ax=ax[0], palette=["#f59e0b", "#4f46e5"])
ax[0].set_title("Students with internet score slightly higher")
sns.kdeplot(data=st, x="study_hours", hue="school_type", fill=True, ax=ax[1], common_norm=False)
ax[1].set_title("Study hours: similar in both school types")
plt.tight_layout(); plt.show()
hue= splits one chart into groups – no manual filtering needed.
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
st = pd.read_csv("students_performance.csv")
order = ["School", "Graduate", "Post-Graduate"]
fig, ax = plt.subplots(1, 2, figsize=(11, 4))
sns.boxplot(data=st, x="parent_education", y="average", order=order, ax=ax[0], color="#a5b4fc")
ax[0].set_title("Box plot: median and spread")
sns.violinplot(data=st, x="parent_education", y="average", order=order, ax=ax[1], color="#99f6e4", inner="quartile")
ax[1].set_title("Violin: the full shape")
plt.tight_layout(); plt.show()
Children of graduate parents have a higher median, but the groups overlap a lot.
Categories with error bars
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
t = pd.read_csv("telecom_churn.csv")
t["churned"] = (t["churn"] == "Yes").astype(int) * 100
fig, ax = plt.subplots(1, 2, figsize=(11, 4))
sns.countplot(data=t, x="contract", hue="churn", order=["Monthly", "One year", "Two year"], ax=ax[0], palette=["#94a3b8", "#dc2626"])
ax[0].set_title("Customers by contract and churn")
sns.barplot(data=t, x="contract", y="churned", hue="internet_service", order=["Monthly", "One year", "Two year"], ax=ax[1], errorbar=("ci", 95))
ax[1].set_ylabel("Churn %"); ax[1].set_title("Churn % with 95% confidence interval")
plt.tight_layout(); plt.show()
barplot shows the mean and a confidence interval – short lines mean we are more certain.
📏 What is the black line on each bar?
It is the 95% confidence interval: the range where the true average probably lies. If two intervals do not overlap, the difference is likely real. Day 29 explains this properly.
Relationships
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
st = pd.read_csv("students_performance.csv")
fig, ax = plt.subplots(1, 2, figsize=(11, 4))
sns.regplot(data=st, x="study_hours", y="average", ax=ax[0], scatter_kws=dict(s=10, alpha=0.4), line_kws=dict(color="#dc2626"))
ax[0].set_title("Each extra study hour ≈ higher average")
sns.regplot(data=st, x="attendance", y="average", ax=ax[1], scatter_kws=dict(s=10, alpha=0.4, color="#0891b2"), line_kws=dict(color="#dc2626"))
ax[1].set_title("Attendance helps too, but less")
plt.tight_layout(); plt.show()
print(st[["study_hours", "attendance", "average"]].corr().round(2))
study_hours attendance average
study_hours 1.00 -0.01 0.79
attendance -0.01 1.00 0.39
average 0.79 0.39 1.00
regplot draws a best-fit line with its uncertainty band.
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
st = pd.read_csv("students_performance.csv")
pv = st.pivot_table(index="parent_education", columns="internet", values="average", aggfunc="mean")
plt.figure(figsize=(5.5, 3.4))
sns.heatmap(pv.loc[["School", "Graduate", "Post-Graduate"]], annot=True, fmt=".1f", cmap="YlGnBu", cbar_kws=dict(label="avg score"))
plt.title("Average score: parent education × internet"); plt.show()
A heatmap of a pivot table shows two factors together.
pairplot and FacetGrid
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
m = pd.read_csv("mall_customers.csv")
g = sns.pairplot(m[["age", "annual_income_lakh", "spending_score"]], corner=True, plot_kws=dict(s=14, alpha=0.6), height=2.4)
g.figure.suptitle("Mall customers – look at income vs spending!", y=1.02)
plt.show()
The income × spending panel shows five clear groups – we will find them with K-Means on Day 46.
Python
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
h = pd.read_csv("house_prices.csv")
g = sns.FacetGrid(h, col="city", col_wrap=3, height=2.6, aspect=1.3)
g.map_dataframe(sns.scatterplot, x="area_sqft", y="price_lakh", hue="furnishing", s=12, palette="Set2")
g.add_legend(); g.set_titles("{col_name}")
plt.show()
Small multiples: the same chart for each city makes comparison easy.
💡 Styling in one line
sns.set_theme(style="whitegrid", palette="muted", font_scale=1.1) at the top of a notebook makes every chart consistent.
🏅 Job skills unlocked today
✔ Plot distributions by group with hue✔ Compare categories with confidence intervals✔ Show relationships with regplot and heatmap✔ Explore many variables with pairplot / FacetGrid
Practice tasks
🏋️ Practice – do it yourself
Task 1: Box plot of monthly_charges by churn in the telecom data. Who pays more – churners or stayers?Show solution
Python
import pandas as pd, seaborn as sns, matplotlib.pyplot as plt
t = pd.read_csv("telecom_churn.csv")
plt.figure(figsize=(5, 3.5))
sns.boxplot(data=t, x="churn", y="monthly_charges", palette=["#94a3b8", "#dc2626"], hue="churn", legend=False)
plt.title("Churners pay higher monthly charges"); plt.show()
print(t.groupby("churn")["monthly_charges"].median())
churn
No 640.0
Yes 790.0
Name: monthly_charges, dtype: float64
Output of the program
Task 2: Heatmap of loan approval rate by employment type and credit-score band (<650, 650–750, >750).Show solution
Task 3: Use sns.lineplot to show monthly average AQI in Delhi with one line per year (hue="year").
Quick quiz
🧠 Quick Quiz – test yourself
1. hue="gender" in Seaborn…
One colour per category.
2. A violin plot shows more than a box plot because it shows…
Width = how many values at that level.
3. sns.barplot plots by default…
Use countplot for counts; estimator= to change.
4. Best chart to scan relationships among 4 numeric columns?
Every pair gets a scatter; diagonals show distributions.
5. Seaborn is built on…
So plt.title, ax.set_xlabel etc. all work.
Interview corner
💼 Interview Corner
How would you show that two groups differ, visually?
Box or violin plots side by side for the distribution, and a barplot of the mean with confidence intervals. Overlapping intervals warn the difference may be noise – confirm with a t-test (Day 30).
When is a heatmap useful?
For a correlation matrix, or for any two-way table (pivot) where you want to spot high and low cells quickly – e.g. churn by contract × internet type.
Summary
Seaborn = DataFrame-aware statistical plots on top of Matplotlib.
histplot/kdeplot/box/violin for distributions; count/barplot for categories.
regplot and heatmap for relationships; pairplot and FacetGrid for many at once.
Confidence intervals show how sure we are.
➡️ Next
Day 23: Plotly – interactive charts you can hover, zoom and share as a web page.
Frequently Asked Questions
Seaborn or Matplotlib – which should I learn?
Both. Seaborn for fast statistical charts, Matplotlib to fine-tune them – every Seaborn chart is a Matplotlib chart underneath.
Why do I get a FutureWarning about palette without hue?
Newer Seaborn wants hue when you pass a palette. Add hue="same column" and legend=False.
