Skip to content
Related Articles
Get the best out of our app
Open App

Related Articles

Implementation of Lasso Regression From Scratch using Python

Improve Article
Save Article
Like Article
Improve Article
Save Article
Like Article


  1. Linear Regression
  2. Gradient Descent


Lasso Regression is also another linear model derived from Linear Regression which shares the same hypothetical function for prediction. The cost function of Linear Regression is represented by J.

\frac{1}{m} \sum_{i=1}^{m}\left(y^{(i)}-h\left(x^{(i)}\right)\right)^{2}

Here, m is the total number of training examples in the dataset.
h(x(i)) represents the hypothetical function for prediction.
y(i) represents the value of target variable for ith training example.

Linear Regression model considers all the features equally relevant for prediction. When there are many features in the dataset and even some of them are not relevant for the predictive model. This makes the model more complex with a too inaccurate prediction on the test set ( or overfitting ). Such a model with high variance does not generalize on the new data. So, Lasso Regression comes for the rescue. It introduced an L1 penalty ( or equal to the absolute value of the magnitude of weights) in the cost function of Linear Regression. The modified cost function for Lasso Regression is given below.

\frac{1}{m}\left[\sum_{i=1}^{m}\left(y^{(i)}-h\left(x^{(i)}\right)\right)^{2}+\lambda \sum_{j=1}^{n} w_{j}\right]

Here, w(j) represents the weight for jth feature.  
n is the number of features in the dataset.
lambda is the regularization strength.

Lasso Regression performs both, variable selection and regularization too.

Mathematical Intuition: 

During gradient descent optimization,  added l1 penalty shrunk weights close to zero or zero.  Those weights which are shrunken to zero eliminates the features present in the hypothetical function. Due to this, irrelevant features don’t participate in the predictive model. This penalization of weights makes the hypothesis more simple which encourages the sparsity ( model with few parameters ).

If the intercept is added, it remains unchanged.

We can control the strength of regularization by hyperparameter lambda. All weights are reduced by the same factor lambda. 

Different cases for tuning values of lambda.

  1. If lambda is set to be 0,   Lasso Regression equals Linear Regression.
  2. If lambda is set to be infinity, all weights are shrunk to zero.

If we increase lambda, bias increases if we decrease the lambda variance increase. As lambda increases, more and more weights are shrunk to zero and eliminates features from the model.


Dataset used in this implementation can be downloaded from the link.

It has 2 columns — “YearsExperience” and “Salary” for 30 employees in a company. So in this, we will train a Lasso Regression model to learn the correlation between the number of years of experience of each employee and their respective salary. Once the model is trained, we will be able to predict the salary of an employee on the basis of his years of experience.



# Importing libraries
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
# Lasso Regression
class LassoRegression() :
    def __init__( self, learning_rate, iterations, l1_penality ) :
        self.learning_rate = learning_rate
        self.iterations = iterations
        self.l1_penality = l1_penality
    # Function for model training
    def fit( self, X, Y ) :
        # no_of_training_examples, no_of_features
        self.m, self.n = X.shape
        # weight initialization
        self.W = np.zeros( self.n )
        self.b = 0
        self.X = X
        self.Y = Y
        # gradient descent learning
        for i in range( self.iterations ) :
        return self
    # Helper function to update weights in gradient descent
    def update_weights( self ) :
        Y_pred = self.predict( self.X )
        # calculate gradients  
        dW = np.zeros( self.n )
        for j in range( self.n ) :
            if self.W[j] > 0 :
                dW[j] = ( - ( 2 * ( self.X[:, j] ).dot( self.Y - Y_pred ) ) 
                         + self.l1_penality ) / self.m
            else :
                dW[j] = ( - ( 2 * ( self.X[:, j] ).dot( self.Y - Y_pred ) ) 
                         - self.l1_penality ) / self.m
        db = - 2 * np.sum( self.Y - Y_pred ) / self.m 
        # update weights
        self.W = self.W - self.learning_rate * dW
        self.b = self.b - self.learning_rate * db
        return self
    # Hypothetical function  h( x ) 
    def predict( self, X ) :
        return self.W ) + self.b
def main() :
    # Importing dataset
    df = pd.read_csv( "salary_data.csv" )
    X = df.iloc[:, :-1].values
    Y = df.iloc[:, 1].values
    # Splitting dataset into train and test set
    X_train, X_test, Y_train, Y_test = train_test_split( X, Y, test_size = 1 / 3, random_state = 0 )
    # Model training
    model = LassoRegression( iterations = 1000, learning_rate = 0.01, l1_penality = 500 ) X_train, Y_train )
    # Prediction on test set
    Y_pred = model.predict( X_test )
    print( "Predicted values ", np.round( Y_pred[:3], 2 ) ) 
    print( "Real values      ", Y_test[:3] )
    print( "Trained W        ", round( model.W[0], 2 ) )
    print( "Trained b        ", round( model.b, 2 ) )
    # Visualization on test set 
    plt.scatter( X_test, Y_test, color = 'blue' )
    plt.plot( X_test, Y_pred, color = 'orange' )
    plt.title( 'Salary vs Experience' )
    plt.xlabel( 'Years of Experience' )
    plt.ylabel( 'Salary' )
if __name__ == "__main__"


Predicted values  [ 40600.91 123294.39  65033.07]
Real values       [ 37731 122391  57081]
Trained W         9396.99
Trained b         26505.43


Note: It automates certain parts of model selection and sometimes called variables eliminator.

My Personal Notes arrow_drop_up
Last Updated : 05 Sep, 2020
Like Article
Save Article
Similar Reads
Related Tutorials