seaborn（Seaborn 前端）

Seaborn - Statistical Data Visualization

Seaborn helps you explore and understand your data through beautiful, informative statistical plots. It automates complex tasks like calculating confidence intervals, aggregating data, and creating faceted grids.

When to Use

• Visualizing complex relationships between multiple variables (relplot)
• Examining univariate and bivariate distributions (displot, kdeplot)
• Comparing categories with statistical summaries (catplot, boxplot, violinplot)
• Visualizing linear regression models and their uncertainty (regplot, lmplot)
• Creating heatmaps and cluster maps for large matrices
• Building multi-plot grids based on data subsets (FacetGrid)
• Setting high-level aesthetic themes for Matplotlib figures

Reference Documentation

Official docs: https://seaborn.pydata.org/ Example gallery: https://seaborn.pydata.org/examples/index.html Search patterns: sns.load_dataset, sns.relplot, sns.catplot, sns.set_theme, sns.heatmap

Core Principles

Figure-Level vs. Axes-Level Functions

Use Seaborn For

• Statistical analysis and exploratory data research (EDA).
• Working directly with Pandas DataFrames in "tidy" (long-form) format.
• Automatic calculation of 95% confidence intervals (error bars).
• Rapidly changing visual themes and color palettes.

Do NOT Use For

• Very low-level custom graphics (use Matplotlib).
• Interactive web visualizations (use Plotly).
• 3D plotting (use Matplotlib mplot3d or PyVista).
• Network graphs (use NetworkX with Matplotlib).

Quick Reference

Installation

pip install seaborn

Standard Imports

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

# Apply the default theme
sns.set_theme()

Basic Pattern - Tidy Data Mapping

import seaborn as sns

# Load an example dataset
tips = sns.load_dataset("tips")

# Create a scatter plot with semantic mapping
sns.relplot(
    data=tips,
    x="total_bill", y="tip",
    hue="smoker", style="time", size="size",
)
plt.show()

Critical Rules

✅ DO

• Use Tidy Data - Ensure your DataFrame is in "long-form" (one row per observation).
• Prefer Figure-Level Functions - Use relplot/displot/catplot for better default layouts and faceting.
• Use the data= parameter - Always pass the DataFrame to keep code clean.
• Set Themes - Use sns.set_theme(style="whitegrid", palette="muted") early in your script.
• Leverage hue - Use semantic color mapping to add extra dimensions to 2D plots.
• Context matters - Use sns.set_context("paper") for publications or "talk" for presentations.

❌ DON'T

• Pass 1D arrays manually - Avoid sns.plot(x_array, y_array); it ignores the power of Pandas integration.
• Ignore the Index - Unlike Matplotlib, Seaborn mostly ignores the DataFrame index (use columns instead).
• Overcrowd plots - Too many semantic mappings (hue, size, style) make graphs unreadable.
• Forget Matplotlib - Remember that Seaborn functions return Matplotlib objects; use ax.set_title() to tweak them.

Anti-Patterns (NEVER)

import seaborn as sns
import matplotlib.pyplot as plt

# ❌ BAD: Iterating through groups to plot manually
for s in df['species'].unique():
    subset = df[df['species'] == s]
    plt.scatter(subset['x'], subset['y'], label=s)

# ✅ GOOD: Let Seaborn handle grouping and legend
sns.scatterplot(data=df, x='x', y='y', hue='species')

# ❌ BAD: Mixing Seaborn and Matplotlib titles incorrectly
sns.displot(data=df, x='val')
plt.title("My Title") # ⚠️ Might apply to the wrong axis in a FacetGrid!

# ✅ GOOD: Use the returned object
g = sns.displot(data=df, x='val')
g.set_axis_labels("Value", "Count")
g.figure.suptitle("Correct Global Title", y=1.05)

Relational Plots (relplot)

Scatter and Line Plots

# Multi-faceted scatter plot
sns.relplot(
    data=tips, x="total_bill", y="tip",
    col="time", hue="day", style="sex",
    kind="scatter"
)

# Line plot with automatic aggregation (mean + 95% CI)
fmri = sns.load_dataset("fmri")
sns.relplot(
    data=fmri, x="timepoint", y="signal",
    hue="event", style="region",
    kind="line", errorbar="sd" # "sd" for standard deviation instead of CI
)

Distribution Plots (displot)

Histograms and KDEs

penguins = sns.load_dataset("penguins")

# Histogram with Kernel Density Estimate
sns.displot(data=penguins, x="flipper_length_mm", hue="species", kde=True)

# Bivariate distribution (Heatmap style)
sns.displot(data=penguins, x="bill_length_mm", y="bill_depth_mm", hue="species", kind="kde")

# Empirical Cumulative Distribution (ECDF)
sns.displot(data=penguins, x="flipper_length_mm", hue="species", kind="ecdf")

Categorical Plots (catplot)

Comparisons and Distribution within categories

# Boxplot (Show quartiles and outliers)
sns.catplot(data=tips, x="day", y="total_bill", kind="box")

# Violin plot (Show density and quartiles)
sns.catplot(data=tips, x="day", y="total_bill", hue="sex", kind="violin", split=True)

# Swarm plot (Show every point without overlap)
sns.catplot(data=tips, x="day", y="total_bill", kind="swarm")

# Bar plot (Show mean and error bars)
sns.catplot(data=tips, x="day", y="total_bill", kind="bar", errorbar=("pi", 95))

Regression Plots

Visualizing Linear Trends

# Simple regression with scatter
sns.regplot(data=tips, x="total_bill", y="tip")

# Faceted regression
sns.lmplot(data=tips, x="total_bill", y="tip", col="smoker", hue="time")

# Logistic regression (for binary data)
sns.lmplot(data=df, x="variable", y="binary_outcome", logistic=True)

Matrix Plots

Heatmaps and Clustering

flights = sns.load_dataset("flights").pivot(index="month", columns="year", values="passengers")

# Heatmap
plt.figure(figsize=(10, 8))
sns.heatmap(flights, annot=True, fmt="d", cmap="YlGnBu")

# Cluster map (Hierarchical clustering)
sns.clustermap(flights, standard_scale=1, cmap="mako")

Grid Objects (Advanced)

Custom Multi-plot Layouts

# JointPlot (Scatter + Marginals)
sns.jointplot(data=penguins, x="bill_length_mm", y="bill_depth_mm", hue="species", kind="kde")

# PairPlot (All-against-all relations)
sns.pairplot(data=penguins, hue="species", corner=True)

# Custom FacetGrid
g = sns.FacetGrid(tips, col="time",  row="sex")
g.map(sns.scatterplot, "total_bill", "tip")

Styling and Aesthetics

Themes and Palettes

# Set overall look
sns.set_style("darkgrid") # white, dark, whitegrid, ticks
sns.set_context("talk")   # paper, notebook, talk, poster

# Custom palettes
sns.set_palette("husl") # Set global palette
my_pal = sns.color_palette("rocket", as_cmap=True) # Get palette as object

# Viewing a palette
sns.palplot(sns.color_palette("Set2"))

Practical Workflows

1. Exploratory Data Analysis (EDA) Pipeline

def initial_eda(df, target_col):
    """Generate basic visual summary of a dataset."""
    # 1. Distribution of target
    sns.displot(data=df, x=target_col, kde=True)

    # 2. Pairwise relations of numeric features
    sns.pairplot(data=df, hue=target_col if df[target_col].nunique() < 10 else None)

    # 3. Correlation heatmap
    plt.figure(figsize=(12, 10))
    sns.heatmap(df.corr(numeric_only=True), annot=True, cmap='coolwarm', fmt=".2f")

# initial_eda(iris, "species")

2. Scientific Result Comparison

def plot_experiment_results(df):
    """Plot results of an experiment with multiple conditions."""
    g = sns.catplot(
        data=df, kind="bar",
        x="condition", y="metric", hue="group",
        palette="viridis", alpha=.6, height=6
    )
    g.despine(left=True)
    g.set_axis_labels("Experimental Condition", "Accuracy (%)")
    g.legend.set_title("User Group")
    return g

3. Time-Series Trends by Category

def plot_trends(df, time_col, val_col, cat_col):
    """Visualizes trends over time with confidence intervals."""
    plt.figure(figsize=(12, 6))
    sns.lineplot(
        data=df, x=time_col, y=val_col, hue=cat_col,
        marker="o", err_style="bars"
    )
    plt.xticks(rotation=45)
    plt.tight_layout()

Common Pitfalls and Solutions

Legend Outside the Plot

# ❌ Problem: Legend covers data in narrow plots
# ✅ Solution: Move legend manually using Matplotlib logic
g = sns.scatterplot(data=tips, x="total_bill", y="tip", hue="day")
sns.move_legend(g, "upper left", bbox_to_anchor=(1, 1))

Slow Performance with Large Data

# ❌ Problem: sns.pairplot(large_df) hangs
# ✅ Solution: Sample data or use simpler plots
sns.pairplot(df.sample(1000), hue='category')
# OR use hist instead of scatter
sns.jointplot(data=df, x='x', y='y', kind="hist")

Overlapping Labels

# ❌ Problem: Categorical labels on X-axis overlap
# ✅ Solution: Rotate labels using Matplotlib
g = sns.boxplot(data=df, x='very_long_category_name', y='value')
g.set_xticklabels(g.get_xticklabels(), rotation=45, horizontalalignment='right')

Best Practices

1. Use tidy data format - Ensure your DataFrame is in long-form (one row per observation)
2. Prefer figure-level functions - Use relplot, displot, and catplot for better default layouts and faceting
3. Always use the data= parameter - Pass the DataFrame directly to keep code clean and readable
4. Set themes early - Use sns.set_theme() at the beginning of your script for consistent styling
5. Leverage semantic mappings - Use hue, size, and style to add dimensions to your plots
6. Choose appropriate context - Use sns.set_context("paper") for publications or "talk" for presentations
7. Remember Seaborn returns Matplotlib objects - Use Matplotlib methods like ax.set_title() for fine-tuning
8. Don't overcrowd plots - Limit semantic mappings to maintain readability
9. Use figure-level functions for faceting - They handle subplot layouts automatically
10. Sample large datasets - Use df.sample() before plotting to improve performance with big data

Seaborn makes statistical visualization a joy by providing high-level abstractions that produce beautiful, publication-quality graphics with minimal effort.