Reward Scaling

Problem

Scale rewards by a running standard deviation.

Return rewards[i] / (std + epsilon) so the scaled rewards have roughly unit scale. An empty list returns [].

Hint

The epsilon keeps the division safe when the running std is near zero.

case 1

in: [[1,2,3],1]

expected: [0.9999999900000002,1.9999999800000003,2.9999999700000006]

case 2

in: [[10],0.5]

expected: [19.999999600000006]

case 3

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

expected: [-0.4999999975,0.4999999975]

case 4

in: [[],1]

expected: []

Starter codePython

def scale_rewards(rewards, std, epsilon=1e-8):
    # Your code here
    pass

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