You will simulate gradient descent on a simple linear regression problem to compare how feature scaling affects convergence speed.

Steps:

1. Generate synthetic data X (one feature) and y using the relation y = 3 * X + noise.
2. Implement a simple gradient descent function that minimizes MSE loss:

   def gradient_descent(X, y, lr, steps):
       w = 0.0
       history = []
       for _ in range(steps):
           grad = -2 * np.mean(X * (y - w * X))
           w -= lr * grad
           history.append(w)
       return np.array(history)
   

3. Run gradient descent twice:
   • on the original X,
   • and on the standardized X_scaled = (X - mean) / std.
4. Plot or print the loss decrease for both to see that scaling accelerates convergence.
5. Compute and print final weights and losses for both cases.