Open In App

Understanding LARS Lasso Regression

Last Updated : 17 Dec, 2023
Like Article

A regularization method called LARS Lasso (Least Angle Regression Lasso) is used in linear regression to decrease the number of features and enhance the model’s predictive ability. It is a variation on the Lasso (Least Absolute Shrinkage and Selection Operator) regression, in which certain regression coefficients shrink to zero as a result of penalizing the absolute values of the regression coefficients. By successfully eliminating unnecessary characteristics from the model, the data is represented in a way that is easier to understand and more economical.

Least Angle Regression (LARS)

A linear regression algorithm called Least Angle Regression (LARS) is intended for high-dimensional data. It effectively calculates a solution path as a function of the regularization parameter, demonstrating how regularization affects the coefficients in the model. The way LARS functions is by repeatedly choosing the predictor that has the strongest correlation with the answer but hasn’t been added to the active set. Following that, it proceeds in the direction that minimizes the angle formed by the residual and the current predictor. This process keeps going until the target number of features is attained. LARS provides a thorough understanding of feature importance and is especially helpful for datasets that have more predictors than observations.

LARS Lasso

The Least Angle Regression (LARS) method The Lasso strategy in linear regression combines the regularization capabilities of L1 regularization (also called Lasso) with the efficiency of forward selection. Loss of arousal As it moves in the direction of the target variable’s greatest correlation, Lasso gradually adds features to the model. This process is continued until a point is reached at which the correlation between the additional variable and the original one would be equal. Due to its tendency to choose a sparse subset of characteristics, this approach works especially well with high-dimensional data. Models with fewer non-zero coefficients are encouraged to be sparse by the regularization term (L1 penalty).

Why LARS Lasso?

In comparison to conventional Lasso regression, LARS Lasso has the following benefits:

  1. Efficiency: For big datasets with plenty of characteristics, LARS Lasso is computationally more efficient than Lasso regression. This is because, instead of tackling a challenging optimization issue, it makes use of an effective method that repeatedly adds the most illuminating characteristic at each phase.
  2. Stability: The LARS Lasso is renowned for its ability to choose features steadily. LARS Lasso offers a consistent feature selection procedure that is less vulnerable to fluctuations in the data, in contrast to Lasso regression, which might be sensitive to the sequence in which features are introduced to the model.
  3. Interpretability: LARS Lasso’s path of coefficient estimates offers important information about the relative significance of various characteristics. The characteristics that most substantially increase the model’s prediction capacity may be found by tracking the coefficients’ changes throughout the regularization process.

Putting LARS Lasso to Use in Python

The well-known Python machine learning module Scikit-Learn offers a practical LARS Lasso implementation. Users may fit the model to the training data and provide the regularization parameter (alpha) using the LassoLarsCV class in Scikit-Learn. The coefficients, which indicate the relative relevance of each characteristic, may then be extracted from the fitted model and used to make predictions on fresh data.

Parameters of LARS Lasso

The LARS Lasso algorithm, a linear model with L1 regularization, is implemented by the LassoLars class in scikit-learn. The parameters are explained as follows:

  • alpha: Strength of regularization. It’s a hyperparameter that must be adjusted according to the particular dataset and problem. More regularization is produced by higher alpha values, which may result in sparser models.
  • fit_intercept: Whether or not to compute this model’s intercept. No intercept will be used in computations if set to False.
  • verbose: Verbosity level. If True, progress information is printed during fitting.
  • normalize: If its true, the regressors will be normalized before regression by subtracting the mean and dividing by the l2-norm.
  • precompute: if using a precomputed Gram matrix to expedite computations is appropriate. If n_samples * n_features < 2^18, then ‘auto’ employs a precomputed Gram matrix. Precomputation can be forced on or disabled by setting it to True or False.
  • max_iter: maximum number of iterations for optimization algorithm.
  • fit_path: If this is the case, the coef_path_ attribute contains the entire path.
  • positive: If it is true, then only positive coefficients are allowed.

Concepts of LARS Lasso

  • L1-Regularization: The linear regression objective function gains a penalty term from LARS Lasso depending on the absolute values of the coefficients. In order to encourage sparsity, this pushes some coefficients to be exactly zero.
  • Regularization strength (alpha): The Lasso and LARS Lasso coefficients’ intensity of penalty is determined by the regularization parameter alpha. A sparser model is produced when the alpha value is higher since it causes the coefficients to shrink more. Cross-validation is usually used to establish the ideal value of alpha.
  • Coefficient Path: As the regularization parameter alpha changes from 0 to 1 (or any other maximum value), LARS Lasso generates a path of coefficient estimates. Given that the coefficients of more significant characteristics tend to vary more gradually along the road, this path provides light on the relative value of the features.
  • Forward Feature Selection: Forward feature selection is carried out by LARS Lasso, which means that it adds predictors to the model one at a time and advances in the direction of the predictor that has the highest correlation with the response at each step. Its efficiency in high-dimensional spaces arises from this.
  • Orthogonal Active Sets: The predictors that have been added to the model are the orthogonal active set of predictors that LARS Lasso keeps track of. It is computationally efficient because of this orthogonalization.

Implementation of LARS Lasso

Import necessary libraries


import numpy as np
from sklearn.linear_model import LassoLars
import matplotlib.pyplot as plt

Generate or load your dataset


# example data generation
X = np.random.rand(100, 10)
y = 2 * X[:, 2] + 1.5 * X[:, 5] + np.random.normal(0, 0.5, 100)

This code creates fictitious data to solve a regression issue. For reproducibility, a random seed is set, a matrix X of shape (100, 10) is created with random values, and the target variable y is produced as a linear combination of X’s third and sixth columns plus additional Gaussian noise.

Apply LARS Lasso


# Apply LARS Lasso with a different alpha value
lars_lasso = LassoLars(alpha=0.05# Adjust alpha as needed, y)

This code uses a regularization parameter (alpha) of 0.05 to apply the LARS Lasso algorithm. The fit method is used to fit the LARS Lasso to the input data (X) and target values (y). The degree of regularization is regulated by the alpha parameter, which can be adjusted by users according to the degree of penalization they want for the model’s coefficients.

Inspect the coefficients


# Inspect the coefficients
print("Coefficients:", lars_lasso.coef_)


Coefficients: [0.         0.         1.38070234 0.         0.         0.74306045
0. 0. 0. 0. ]

The coefficients of the features that the LARS Lasso regression model learned are printed by this code. Each feature’s contribution to the linear relationship with the target variable is shown by the coefficients. One can determine the significance and effect of each feature in the regression model by looking at these coefficients.

Plot the results


# Plot the results with different labels
plt.plot(lars_lasso.coef_, marker='o', label='Updated LARS Lasso Coefficients')
plt.xlabel('Coefficient Index')
plt.ylabel('Coefficient Value')
plt.title('LARS Lasso Regression with Different Values')



Lasso regression, this code creates a plot. Each coefficient’s index is represented by the x-axis, while the y-axis displays the coefficients’ corresponding values. To draw attention to the coefficient values, use the marker ‘o’.

The LARS Lasso regression’s coefficients on a synthetic dataset are shown in the output graphic. The coefficient of each characteristic is represented by a point on the plot, and its index is shown on the x-axis. The model’s chosen features are shown by the non-zero coefficients, showcasing LARS Lasso’s capacity to pick variables. Both the regularization intensity (alpha) and the underlying data production procedure have an impact on the coefficient values. This image helps with feature selection and understanding by illuminating the significance and influence of each variable in the regression model.

LARS Lasso Applications

Applications for LARS Lasso may be found in a number of fields, including:

  • Feature Selection: To choose the subset of characteristics that are most relevant to the predicting job, LARS Lasso is often used for feature selection in high-dimensional datasets.
  • Sparsity Induction: LARS Lasso encourages sparsity in the model by decreasing unnecessary coefficients to zero, making it appropriate for applications where the real underlying structure of the data is thought to be sparse.
  • Variable relevance Analysis: The relative relevance of various model characteristics may be evaluated using the path of coefficient estimates generated by LARS Lasso. Understanding the underlying correlations between variables and figuring out the main factors influencing the prediction performance may be done with the use of this information.


For linear regression feature selection, sparsity induction, and variable significance analysis, LARS Lasso is a strong and adaptable tool. It is a useful tool for data scientists and machine learning practitioners to have in their toolbox because of its stability, interpretability, and computational efficiency. LARS Lasso offers insights into the structure and connections inside complicated datasets and is easily applied to a broad variety of applications because to its straightforward implementation in Scikit-Learn.

Similar Reads

Least Angle Regression (LARS)
Regression is a supervised machine learning task that can predict continuous values (real numbers), as compared to classification, that can predict categorical or discrete values. Before we begin, if you are a beginner, I highly recommend this article. Least Angle Regression (LARS) is an algorithm used in regression for high dimensional data (i.e.,
3 min read
Multi-task lasso regression
MultiTaskLasso Regression is an enhanced version of Lasso regression. MultiTaskLasso is a model provided by sklearn that is used for multiple regression problems to work together by estimating their sparse coefficients. There is the same feature for all the regression problems called tasks. This model is trained with a mixed l1/l2 norm for regulari
2 min read
What is lasso regression?
The Lasso Regression, a regression method based on Least Absolute Shrinkage and Selection Operator is quite an important technique in regression analysis for selecting the variables and regularization. It gets rid of irrelevant data features that help to prevent overfitting and features with weak influence become more cleanly identifiable because o
10 min read
Implementation of Lasso Regression From Scratch using Python
For regularization and feature selection, Lasso Regression, also known as the Least Absolute Shrinkage and Selection Operator, is a linear regression technique. The cost function of linear regression, which is the sum of squared discrepancies between predicted and actual values, gains a penalty element. The absolute value of the coefficients multip
11 min read
Lasso Regression in R Programming
Lasso regression is a classification algorithm that uses shrinkage in simple and sparse models(i.e models with fewer parameters). In Shrinkage, data values are shrunk towards a central point like the mean. Lasso regression is a regularized regression algorithm that performs L1 regularization which adds a penalty equal to the absolute value of the m
11 min read
Weighted Lasso Regression in R
In the world of data analysis and prediction, regression techniques are essential for understanding relationships between variables and making accurate forecasts. One standout method among many is Lasso regression. It not only helps in finding these relationships but also aids in creating models that are easier to interpret and more resilient. Howe
6 min read
Lasso vs Ridge vs Elastic Net | ML
Bias: Biases are the underlying assumptions that are made by data to simplify the target function. Bias does help us generalize the data better and make the model less sensitive to single data points. It also decreases the training time because of the decrease in complexity of target function High bias suggest that there is more assumption taken on
6 min read
Implementation of Lasso, Ridge and Elastic Net
In this article, we will look into the implementation of different regularization techniques. First, we will start with multiple linear regression. For that, we require the python3 environment with sci-kit learn and pandas preinstall. We can also use google collaboratory or any other jupyter notebook environment.First, we need to import some packag
7 min read
Joint Feature Selection with multi-task Lasso in Scikit Learn
This article is likely to introduces the concepts of Lasso and multitask Lasso regression and demonstrates how to implement these methods in Python using the scikit-learn library. The article is going to cover the differences between Lasso and multitask Lasso and provide guidance on which method to prefer in different situations. It may also provid
7 min read
Comparison between L1-LASSO and Linear SVM
Within machine learning, linear Support Vector Machines (SVM) and L1-regularized Least Absolute Shrinkage and Selection Operator (LASSO) regression are powerful methods for classification and regression, respectively. Although the goal of both approaches is to locate a linear decision boundary, they differ in their features and optimization goals.
3 min read