Seaborn 101 — Pretty Plots Fast

• Install seaborn; load a built-in dataset.
• Draw histplot, countplot, scatterplot, boxplot in one line each.
• Colour points by a category with hue=.
• Read each chart for an ML insight (separability, imbalance, outliers).

pip install seaborn

import seaborn as sns
import matplotlib.pyplot as plt

penguins = sns.load_dataset("penguins")   # built-in sample
print(penguins.head())
print(penguins["species"].value_counts())

histplot — distribution of one number

sns.histplot(data=penguins, x="body_mass_g", bins=30, kde=True)
plt.show()

Shows the shape — is it bell-shaped, skewed, bimodal? kde=True overlays a smooth curve.

countplot — frequency of a category

sns.countplot(data=penguins, x="species")
plt.show()

Instantly reveals class balance. Three roughly equal bars = balanced; one tiny bar = imbalance you must handle.

scatterplot — relationship of two numbers, coloured by class

sns.scatterplot(data=penguins, x="bill_length_mm",
                y="flipper_length_mm", hue="species")
plt.show()

The most useful ML chart. If colours form separate clusters, the classes are separable — a classifier will do well. If they're a jumbled mess, expect trouble.

boxplot — distribution per category, with outliers

sns.boxplot(data=penguins, x="species", y="body_mass_g")
plt.show()

Box = middle 50%, line = median, dots = outliers. Compares groups and flags extremes.

Make it look good

sns.set_theme(style="whitegrid")    # call once at the top
plt.figure(figsize=(8, 5))
# ... your plot ...
plt.title("Penguin body mass by species")
plt.tight_layout()

# explore.py — a 4-panel "should I even train?" check
import seaborn as sns
import matplotlib.pyplot as plt

sns.set_theme(style="whitegrid")
df = sns.load_dataset("penguins").dropna()

fig, axes = plt.subplots(2, 2, figsize=(11, 8))

# class balance
sns.countplot(data=df, x="species", ax=axes[0, 0])
axes[0, 0].set_title("Class balance")

# one feature's distribution
sns.histplot(data=df, x="flipper_length_mm", hue="species",
             kde=True, ax=axes[0, 1])
axes[0, 1].set_title("Flipper length distribution")

# separability
sns.scatterplot(data=df, x="bill_length_mm", y="flipper_length_mm",
                hue="species", ax=axes[1, 0])
axes[1, 0].set_title("Are classes separable?")

# per-class spread
sns.boxplot(data=df, x="species", y="body_mass_g", ax=axes[1, 1])
axes[1, 1].set_title("Body mass by species")

fig.tight_layout()
fig.savefig("explore.png", dpi=150)
plt.show()

Read the diff

Four panels answer four ML questions: balance (roughly even — good), distribution (clear peaks per species), separability (the scatter shows three loose clusters — a classifier will work), spread (no wild outliers). You now expect a good model before training one. That intuition is what separates careful ML practitioners from button-pushers.

Seaborn draws statistical charts in one line from a DataFrame. histplot for distributions, countplot for balance, scatterplot + hue for separability, boxplot for per-group spread. Always explore before training — charts tell you whether a problem is learnable. Tomorrow: the three deeper seaborn plots.