Deep Double Descent: The Ridge Solution

The problem, from first principles

Ridge regression adds an L2 penalty, which makes the normal equations solvable even when there are more features than samples. As the penalty goes to zero, ridge approaches the minimum-norm solution, linking it back to double descent.

From theory to code

Implement ridge_solution(X, y, lam), returning (X^T X + lam I)^{-1} X^T y.

Constraints

• Use np.linalg.solve; do not form an inverse explicitly.

Hints