DQN: The Bellman Target
Beginner
The problem, from first principles
DQN (Mnih et al., 2013/2015) learns an action-value function by regressing it toward a bootstrapped target: the reward plus the discounted best value at the next state. The target is what the Q-network is pulled toward on every update.
From theory to code
Implement td_target(r, gamma, max_next_q, done), returning the one-step Bellman target.
Constraints
- Terminal transitions do not bootstrap.
Hints
Hint 1
Multiply the next-state value by gamma and by (1 - done), then add the reward.
Keywords
research-papersreinforcement-learningdeep-q-learning
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.