Numerical gradients and a descent loop

Build the two tools you will use to check and train every model in this course.

• numerical_gradient(f, x, h=1e-5): given a function f that takes a 1-D array and returns a number, return an array the same shape as x holding the central-difference estimate of each partial derivative. Nudge one element at a time.
• gradient_descent(grad, x0, lr, steps): start at x0 and apply x = x - lr * grad(x) steps times. Return the final x. Don't modify x0 in place.