Reinforcement Learning Algorithms with Python

DQN implementation

Though DQN is a pretty simple algorithm, it requires particular attention when it comes to its implementation and design choices. This algorithm, like every other deep RL algorithm, is not easy to debug and tune. Therefore, throughout this book, we'll give you some techniques and suggestions for how to do this.

The DQN code contains four main components:

• DNNs
• An experienced buffer
• A computational graph
• A training (and evaluation) loop

The code, as usual, is written in Python and TensorFlow, and we'll use TensorBoard to visualize the training and the performance of the algorithm.

All the code is available in this book's GitHub repository. Make sure to check it out there. We don't provide the implementation of some simpler ...