Deriva — Derive the Algorithm

← Projectstensor-autograd / L4Backward Pass~24 min

one move: bound a failure

Replace hand-derived differentiation with a numerical gradient check for f(x)=x². This is the smallest useful test for an autodiff engine. In the Tensor/Autograd Engine system, implement this level as a deterministic contract before composing it with the next service.

Required API

def finite_difference_gradient(x, epsilon):
    return estimated_gradient

Behavior

evaluate the function on both sides of x
divide the symmetric difference by 2*epsilon
round only the reported diagnostic

Examples — tap to reveal

Constraints

Use only the Python standard library.
Return deterministic values so every run can be compared with the contract.
Do not mutate caller-owned inputs.

Opening…