N-Step Window Return

Problem

Compute the discounted return inside a window of an episode.

Starting at index start, sum gamma^k * rewards[start + k] for the first n available rewards, truncating at the end of the episode. Return 0.0 when start is out of range or n is 0.

Hint

This is the raw n-step reward sum with no bootstrap term.

Test cases

case 1

in: [[0,1,2,3,4],1,3,0.5]

expected: 2.75

case 2

in: [[1,2,3],0,5,0.9]

expected: 5.23

case 3

in: [[1,2],5,2,0.9]

expected: 0

case 4

in: [[2,4],0,0,0.9]

expected: 0

Starter codePython

def n_step_window_return(rewards, start, n, gamma):
    # Your code here
    pass

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