KL Early-Stop Check

Problem

Decide whether training should stop early because the policy moved too far, measured by KL(p || q).

Compute KL(p || q) = sum of p * log(p / q), skipping zero probabilities, and return True when it exceeds target_kl.

Hint

The check is strict: a KL exactly equal to target_kl does not trigger a stop.

case 1

in: [[0.5,0.5],[0.5,0.5],0.01]

expected: false

case 2

in: [[0.9,0.1],[0.5,0.5],0.1]

expected: true

case 3

in: [[1,0],[0.5,0.5],0.5]

expected: true

case 4

in: [[0.5,0.5],[0.5,0.5],0]

expected: false

Starter codePython

import math
def kl_early_stop(p, q, target_kl):
    # Your code here
    pass

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