Implementation:Scikit learn Scikit learn Lasso

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Machine Learning, Sparse Regression
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for linear regression with L1 regularization (Lasso) enabling sparse feature selection provided by scikit-learn.

Description

Lasso (Least Absolute Shrinkage and Selection Operator) implements a linear model trained with L1 prior as regularizer. It extends ElasticNet with l1_ratio fixed at 1.0. The optimization objective minimizes (1 / (2 * n_samples)) * ||y - Xw||^2_2 + alpha * ||w||_1. The L1 penalty encourages sparsity in the learned coefficients, effectively performing feature selection by driving some coefficients exactly to zero. The implementation uses coordinate descent optimization.

Usage

Use Lasso when you need a linear regression model with automatic feature selection, when you suspect many features are irrelevant, or when you want an interpretable sparse model. It is especially useful in high-dimensional settings where the number of features may exceed the number of samples.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/linear_model/_coordinate_descent.py

Signature

class Lasso(ElasticNet):
    def __init__(
        self,
        alpha=1.0,
        *,
        fit_intercept=True,
        precompute=False,
        copy_X=True,
        max_iter=1000,
        tol=1e-4,
        warm_start=False,
        positive=False,
        random_state=None,
        selection="cyclic",
    ):

Import

from sklearn.linear_model import Lasso

I/O Contract

Inputs

Name	Type	Required	Description
alpha	float	No	Constant multiplying the L1 penalty term (default=1.0)
fit_intercept	bool	No	Whether to calculate the intercept (default=True)
precompute	bool or array-like	No	Whether to use a precomputed Gram matrix (default=False)
copy_X	bool	No	Whether to copy X or overwrite (default=True)
max_iter	int	No	Maximum number of iterations (default=1000)
tol	float	No	Tolerance for the optimization convergence (default=1e-4)
warm_start	bool	No	Reuse solution of previous call as initialization (default=False)
positive	bool	No	Restrict coefficients to be positive (default=False)
random_state	int or RandomState	No	Random number generator seed for selection='random'
selection	str	No	Coefficient update order: 'cyclic' or 'random' (default='cyclic')

Outputs

Name	Type	Description
coef_	ndarray of shape (n_features,)	Estimated coefficients for the linear model
intercept_	float	Estimated independent term in the linear model
n_iter_	int	Number of iterations run by the coordinate descent solver
dual_gap_	float	The dual gap at the end of optimization

Usage Examples

Basic Usage

from sklearn.linear_model import Lasso
from sklearn.datasets import make_regression

X, y = make_regression(n_samples=100, n_features=20, n_informative=5, noise=10, random_state=42)
model = Lasso(alpha=0.1)
model.fit(X, y)
print("Non-zero coefficients:", (model.coef_ != 0).sum())
print("Score:", model.score(X, y))

Related Pages

• Principle:Scikit_learn_Scikit_learn_Linear_Regression