seaborn

seaborn Best Practices

Seaborn is the definitive library for statistical data visualization in Python. This guide outlines our team's mandatory best practices to ensure consistent, reproducible, and high-quality plots in all AI/ML projects.

1. Code Organization & Project-Wide Styling

Always configure Seaborn's global style settings once per project. Centralize this in a dedicated styles.py module and import it. This guarantees a uniform visual language across all outputs.

❌ BAD: Ad-hoc styling in every script/notebook

# my_script.py
import seaborn as sns
import matplotlib.pyplot as plt

sns.set_theme(style="darkgrid", palette="viridis") # In every file!
# ... plotting code

✅ GOOD: Centralized and imported styling

# styles.py
import seaborn as sns
import matplotlib.pyplot as plt

def apply_seaborn_defaults():
    """Applies project-wide Seaborn and Matplotlib styling."""
    sns.set_theme(
        context="talk",      # Readable text for presentations/reports
        style="whitegrid",   # Clean background with light grid
        palette="deep"       # Perceptually uniform, colorblind-friendly
    )
    plt.rcParams["figure.figsize"] = (8, 5)    # Consistent figure size
    plt.rcParams["figure.dpi"] = 150           # On-screen clarity

# my_script.py
import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
from .styles import apply_seaborn_defaults # Adjust import path as needed

apply_seaborn_defaults()
# ... plotting code

2. Prefer Figure-Level Functions

For complex layouts, especially with faceting, always use Seaborn's figure-level functions (relplot, catplot, pairplot, lmplot). They handle figure creation, axis management, and legends automatically, ensuring consistency.

❌ BAD: Manual subplots with axes-level functions for faceting

# Hard to manage multiple subplots, shared axes, and legends manually
fig, axes = plt.subplots(1, 2, figsize=(12, 5))
sns.scatterplot(data=df[df['category'] == 'A'], x='x', y='y', ax=axes[0])
sns.scatterplot(data=df[df['category'] == 'B'], x='x', y='y', ax=axes[1])
plt.tight_layout()
plt.show()