Join Our 5-Week ML/AI Engineer Interview Bootcamp 🚀 led by ML Tech Leads at FAANGs
9. Linear Regression
GoogleImplement linear regression from scratch using NumPy, trained with batch gradient descent to minimize mean squared error. Provide fit to learn weights and bias and predict to output continuous values.
Requirements
Implement a class LinearRegressionGD:
Rules:
- Optimize Mean Squared Error (MSE) using batch gradient descent.
- Parameters: weights vector
wand biasb; do not add an intercept column to X. __init__must set defaults and initialize params (e.g.,self.w = np.array([]),self.b = 0.0).fitlearnswandb;predictreturnsX @ w + b.- Use NumPy only with vectorized operations.
Default hyperparameters:
learning_rate = 0.1n_iters = 1000
Example
Output:
| Argument | Type |
|---|---|
| X_test | list |
| X_train | list |
| y_train | list |
| Return Name | Type |
|---|---|
| value | list |
Constraints
NumPy only; no sklearn/statsmodels
Vectorized batch GD; no per-sample loops
No intercept column; use separate bias b
Hint 1
Start by converting X to a 2D np.ndarray and y to a 1D array (y.reshape(-1)), then initialize w with zeros of length n_features and b = 0.0.
Hint 2
Use a fully vectorized forward pass each iteration: y_pred = X @ w + b and error = y_pred - y. Batch gradient descent means compute gradients using all samples each step (no loops over rows).
Hint 3
For MSE, the gradients are dw = (1/n) * (X.T @ error) and db = (1/n) * error.sum(). Update with w -= lr * dw, b -= lr * db, and ensure predict(X) returns X @ w + b without adding an intercept column.
Roles
Companies
Google
AmazonLevels
Tags
Python Editor