Course contentsShow
← Machine Learning and Deep Learning
Lesson 392 of 3,5389. Dimensionality ReductionFree lesson

PCA Implementation in scikit-learn

Using sklearn.decomposition.PCA: fitting, transforming, and accessing components and explained variance.

PCA Implementation in scikit-learn

What You'll Learn

You'll learn how to use scikit-learn's PCA class to perform dimensionality reduction by fitting, transforming data, and extracting components and variance information.

The Core Workflow

Now that you understand PCA theory, scikit-learn makes implementation straightforward with the PCA class from sklearn.decomposition. The workflow follows the familiar fit-transform pattern you've seen with other sklearn models.

Creating and Fitting PCA

First, create a PCA object specifying how many components you want:

from sklearn.decomposition import PCA

# Create PCA with 2 components
pca = PCA(n_components=2)

# Fit to your standardized data
pca.fit(X_standardized)

The fit() method computes the covariance matrix, performs eigendecomposition, and identifies the principal components.

Transforming Data

Transform your original data into the new coordinate system:

# Transform to reduced dimensions
X_pca = pca.transform(X_standardized)

# Or fit and transform in one step
X_pca = pca.fit_transform(X_standardized)

Accessing Results

After fitting, you can inspect what PCA learned:

# The principal components (eigenvectors)
components = pca.components_  # shape: (n_components, n_features)

# Explained variance by each component
explained_var = pca.explained_variance_

# Explained variance ratio (proportion of total variance)
explained_var_ratio = pca.explained_variance_ratio_

# Cumulative explained variance
import numpy as np
cumulative_var = np.cumsum(explained_var_ratio)

The explained_variance_ratio_ is particularly useful for determining how much information each component retains—values closer to 1.0 mean that component captures most of the original variance.

Key Takeaway: Scikit-learn's PCA class provides a simple fit-transform interface to reduce dimensionality, with components_ giving you the new axes and explained_variance_ratio_ showing how much information each component preserves.