Custom Loss for Gradient Descent | Dataford Interview Questions

Code a custom loss function for a gradient descent optimization problem from scratch.

Implement batch gradient descent for linear regression using the Huber loss. Do not use NumPy, machine learning libraries, or automatic differentiation. Return the learned weight vector, including the intercept as its first element.

The function receives feature rows, targets, learning rate, epoch count, and Huber threshold. Initialize all weights to zero and average gradients across each batch.

Constraints

1 <= len(X) <= 1000
1 <= len(X[0]) <= 20
len(X) == len(y)
All feature rows have the same length
0 <= epochs <= 10000
learning_rate > 0
delta > 0

Function Signature

def fit_huber_regression(X, y, learning_rate, epochs, delta):

Your question is Custom Loss for Gradient Descent. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem

Code a custom loss function for a gradient descent optimization problem from scratch.

The function receives feature rows, targets, learning rate, epoch count, and Huber threshold. Initialize all weights to zero and average gradients across each batch.

Constraints

1 <= len(X) <= 1000
1 <= len(X[0]) <= 20
len(X) == len(y)
All feature rows have the same length
0 <= epochs <= 10000
learning_rate > 0
delta > 0

Function Signature

def fit_huber_regression(X, y, learning_rate, epochs, delta):

Your question is Custom Loss for Gradient Descent. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.