Fletcher-Reeves Beta With Restart

Problem

Compute the Fletcher-Reeves conjugate gradient coefficient with a restart test.

Return (grad dot grad) / (grad_old dot grad_old), but return 0.0 when the old gradient is zero or when grad dot p_old is not negative, meaning the previous direction is no longer a descent direction.

Hint

A non-negative gradient-direction dot product triggers a restart to steepest descent.

Test cases

case 1

in: [[2,-2],[1,1],[-1,0]]

expected: 4

case 2

in: [[1,0],[1,1],[1,0]]

expected: 0

case 3

in: [[0,0],[0,0],[1,1]]

expected: 0

case 4

in: [[1,2],[0.5,0.5],[-2,-4]]

expected: 10

Starter codePython

def fletcher_reeves_beta(grad, grad_old, p_old):
    # Your code here
    pass

4 test cases · run your code against them in the editor