A minimal, educational implementation of deep Q-learning methods for the Atari Breakout environment. The goal is to illustrate how to build and Atari agent from scratch. In particular this proejct implements:
- Deep Q-Networks, now famous approach to solving Atari games directly from raw pixels
- Double DQN, reducing overestimation in value estimates
- Prioritized Experience Replay, giving more weight to transitions with high temporal-difference errors
If you are interested, I wrote a small post walking through the implementation.
- Playing Atari with Deep Reinforcement Learning (Mnih et al., 2013)
- Deep Reinforcement Learning with Double Q-learning (Van Hasselt et al., 2015)
- Prioritized Experience Replay (Schaul et al., 2016)