Hands-On Reinforcement Learning with Python

Chapter 9

1. DRQN makes use of recurrent neural network (RNN) where DQN makes use of vanilla neural network.
2. DQN is not used applied when the MDP is partially observable.
3. Refer section Doom with DRQN.
4. DARQN makes use of attention mechanism unlike DRQN.
5. DARQN is used to understand and focus on particular area of game screen which is more important.
6. Soft and hard attention.
7. We set living reward to 0 which the agent does for each move, even though the move is not useful.