Skip to main content

Seaborn: Statistical Visualization

Seaborn is a Python data visualization library based on Matplotlib. It provides a high-level interface for drawing attractive and informative statistical graphics. In Machine Learning, we use Seaborn to quickly understand the distribution of our features and the correlations between them.

1. Why Seaborn over Matplotlib?​

Seaborn is designed to work directly with Pandas DataFrames. It automates much of the boilerplate code (like labeling axes and handling colors) that Matplotlib requires.

2. Visualizing Distributions​

Before training a model, you need to know if your data is "Normal" (Gaussian) or skewed.

Histograms and KDEs​

The displot() (distribution plot) combines a histogram with a Kernel Density Estimate (KDE) to show the "shape" of your data.

import seaborn as sns
import matplotlib.pyplot as plt

# Visualizing the distribution of a feature
sns.displot(df['feature_name'], kde=True, color="skyblue")
plt.show()

3. Visualizing Relationships​

Scatter Plots and Regressions​

In ML, we often want to see if one feature can predict another. regplot() draws a scatter plot and fits a Linear Regression line automatically.

# Check for linear relationship between 'SquareFootage' and 'Price'
sns.regplot(data=df, x="SquareFootage", y="Price")

The Pair Plot​

The pairplot() is perhaps the most useful tool in EDA. It creates a matrix of plots, showing every feature's relationship with every other feature.

# Instant overview of the entire dataset
sns.pairplot(df, hue="Target_Class")

4. Visualizing Categorical Data​

When dealing with discrete categories (like "City" or "Product Type"), we use plots that show the central tendency and variance.

  • Box Plot: Shows the median, quartiles, and outliers.
  • Violin Plot: Combines a box plot with a KDE to show the density of the data at different values.
# Comparing distribution across categories
sns.boxplot(data=df, x="Category", y="Value")

5. Matrix Plots: Correlation Analysis​

Before selecting features for your model, you must check for Multicollinearity (features that are too similar to each other). We do this using a Correlation Matrix visualized as a Heatmap.

# Compute correlation matrix
corr = df.corr()

# Visualize with Heatmap
sns.heatmap(corr, annot=True, cmap='coolwarm', fmt=".2f")

6. Seaborn "Themes" and Aesthetics​

Seaborn makes it easy to change the look of your plots globally to match a professional report or dark-mode dashboard.

  • sns.set_theme(style="whitegrid")
  • sns.set_context("talk") (Scales labels for presentations)

References for More Details​

You have now mastered the "Big Three" of Python data science: NumPy for math, Pandas for data, and Matplotlib/Seaborn for sight. You are ready to stop preparing and start predicting.