Implementation:Scikit learn Scikit learn SparsePCA

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Dimensionality Reduction, Sparse Coding
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for Sparse Principal Components Analysis provided by scikit-learn.

Description

SparsePCA finds the set of sparse components that can optimally reconstruct the data, combining the ability of PCA to extract the most important directions with L1 sparsity on the components. The amount of sparseness is controllable by the alpha parameter. Internally it uses dictionary learning to extract sparse components. The module also provides MiniBatchSparsePCA for a faster but less accurate variant using mini-batch dictionary learning.

Usage

Use SparsePCA when you need interpretable principal components with many zero loadings. It is ideal for applications where you want to understand which features contribute to each component, such as in genomics, neuroimaging, and financial data analysis where sparse, interpretable factors are preferred over dense PCA components.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/decomposition/_sparse_pca.py

Signature

class SparsePCA(_BaseSparsePCA):
    def __init__(
        self,
        n_components=None,
        *,
        alpha=1,
        ridge_alpha=0.01,
        max_iter=1000,
        tol=1e-8,
        method="lars",
        n_jobs=None,
        U_init=None,
        V_init=None,
        verbose=False,
        random_state=None,
    ):

Import

from sklearn.decomposition import SparsePCA

I/O Contract

Inputs

Name	Type	Required	Description
n_components	int	No	Number of sparse atoms to extract. Defaults to n_features if None.
alpha	float	No	Sparsity controlling parameter. Higher values lead to sparser components (default=1).
ridge_alpha	float	No	Amount of ridge shrinkage for the transform method (default=0.01).
max_iter	int	No	Maximum number of iterations (default=1000).
tol	float	No	Tolerance for the stopping condition (default=1e-8).
method	str	No	Optimization method: 'lars' or 'cd' (default='lars').
n_jobs	int	No	Number of parallel jobs to run.
U_init	ndarray	No	Initial values for loadings (warm restart).
V_init	ndarray	No	Initial values for components (warm restart).
verbose	int or bool	No	Controls verbosity (default=False).
random_state	int or RandomState	No	Random state for reproducibility.

Outputs

Name	Type	Description
components_	ndarray of shape (n_components, n_features)	Sparse components extracted from the data.
error_	ndarray	Vector of errors at each iteration.
n_components_	int	Estimated number of components.
n_iter_	int	Number of iterations run.
mean_	ndarray of shape (n_features,)	Per-feature empirical mean.
n_features_in_	int	Number of features seen during fit.

Usage Examples

Basic Usage

import numpy as np
from sklearn.decomposition import SparsePCA

X = np.random.rand(20, 10)
spca = SparsePCA(n_components=5, alpha=1, random_state=0)
X_transformed = spca.fit_transform(X)
print(X_transformed.shape)  # (20, 5)
print(spca.components_.shape)  # (5, 10)
# Components are sparse (many zeros)
print(np.sum(spca.components_ == 0))

Related Pages

• Principle:Scikit_learn_Scikit_learn_Dimensionality_Reduction