Implementation of Elastic Net Regression From Scratch

Last Updated : 21 Mar, 2024

Prerequisites:

Linear Regression
Gradient Descent
Lasso & Ridge Regression

Introduction:

Elastic-Net Regression is a modification of Linear Regression which shares the same hypothetical function for prediction. The cost function of Linear Regression is represented by J.

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

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 i^th training example.

Linear Regression suffers from overfitting and can’t deal with collinear data. 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, to deal with these issues, we include both L-2 and L-1 norm regularization to get the benefits of both Ridge and Lasso at the same time. The resultant model has better predictive power than Lasso. It performs feature selection and also makes the hypothesis simpler. The modified cost function for Elastic-Net Regression is given below :

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

Here, w_(j) represents the weight for j^th feature. n is the number of features in the dataset. lambda1 is the regularization strength for L-1 norm. lambda2 is the regularization strength for L-2 norm.

Mathematical Intuition:

During gradient descent optimization of its cost function, added L-2 penalty term leads to reduces the weights of the model close to zero. Due to the penalization of weights, the hypothesis gets simpler, more generalized, and less prone to overfitting. 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 predictive which encourages the sparsity ( model with few parameters ).

Different cases for tuning values of lambda1 and lamda2.

If lambda1 and lambda2 are set to be 0, Elastic-Net Regression equals Linear Regression.
If lambda1 is set to be 0, Elastic-Net Regression equals Ridge Regression.
If lambda2 is set to be 0, Elastic-Net Regression equals Lasso Regression.
If lambda1 and lambda2 are set to be infinity, all weights are shrunk to zero

So, we should set lambda1 and lambda2 somewhere in between 0 and infinity.

Implementation:

It has 2 columns — “YearsExperience” and “Salary” for 30 employees in a company. So in this, we will train an Elastic-Net 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.

Code:

# Importing libraries 
  
import numpy as np 
  
import pandas as pd 
  
from sklearn.model_selection import train_test_split 
  
import matplotlib.pyplot as plt 
  
# Elastic Net Regression 
  
class ElasticRegression() : 
      
    def __init__( self, learning_rate, iterations, l1_penality, l2_penality ) : 
          
        self.learning_rate = learning_rate 
          
        self.iterations = iterations 
          
        self.l1_penality = l1_penality 
          
        self.l2_penality = l2_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 ) : 
              
            self.update_weights() 
              
        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 + 2 * self.l2_penality * self.W[j] ) / self.m 
          
            else : 
                  
                dW[j] = ( - ( 2 * ( self.X[:,j] ).dot( self.Y - Y_pred ) )  
                           
                         - self.l1_penality + 2 * self.l2_penality * self.W[j] ) / 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 X.dot( self.W ) + self.b 
          
# Driver Code 
  
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 = ElasticRegression( iterations = 1000,  
                                
                      learning_rate = 0.01, l1_penality = 500, l2_penality = 1 ) 
  
    model.fit( 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' ) 
      
    plt.show() 
        
if __name__ == "__main__" :  
      
    main()

Output:

Predicted values [ 40837.61 122887.43 65079.6 ] Real values [ 37731 122391 57081] Trained W 9323.84 Trained b 26851.84

Elastic-Net Model Visualization

Note: Elastic-Net Regression automates certain parts of model selection and leads to dimensionality reduction which makes it a computationally efficient model.

Implementation of Elastic Net Regression From Scratch

mohitbaliyan

News

Improve

Article Tags :

Practice Tags :

Similar Reads

Machine Learning Algorithms

Machine learning algorithms are essentially sets of instructions that allow computers to learn from data, make predictions, and improve their performance over time without being explicitly programmed. Machine learning algorithms are broadly categorized into three types: Supervised Learning: Algorith

Top 15 Machine Learning Algorithms Every Data Scientist Should Know in 2024

Machine Learning (ML) Algorithms are the backbone of everything from Netflix recommendations to fraud detection in financial institutions. These algorithms form the core of intelligent systems, empowering organizations to analyze patterns, predict outcomes, and automate decision-making processes. Wi

ML | Stochastic Gradient Descent (SGD)

Gradient Descent is an iterative optimization process that searches for an objective function's optimum value (Minimum/Maximum). It is one of the most used methods for changing a model's parameters in order to reduce a cost function in machine learning projects. The primary goal of gradient descen

Asynchronous Advantage Actor Critic (A3C) algorithm

The Asynchronous Advantage Actor Critic (A3C) algorithm is one of the newest algorithms to be developed under the field of Deep Reinforcement Learning Algorithms. This algorithm was developed by Google's DeepMind which is the Artificial Intelligence division of Google. This algorithm was first menti

翻译：