Double DQN Target

Problem

Compute the Double DQN bootstrap target.

If the episode is done, return reward. Otherwise select the greedy next action using online_next (smallest index on ties) and return reward + gamma * target_next[that action].

Hint

The online network picks the action; the target network supplies the value.

Test cases

case 1

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

expected: 1.9

case 2

in: [2,true,0.9,[1,2],[3,4]]

expected: 2

case 3

in: [0,false,1,[2,2,1],[1,9,5]]

expected: 1

case 4

in: [-1,false,0.5,[0,0],[-2,-4]]

expected: -2

Starter codePython

def double_dqn_target(reward, done, gamma, online_next, target_next):
    # Your code here
    pass

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