The algorithms were implemented using the book: "Deep Reinforcement Learning Hands-On" written by Maxim Lapan. He provides a github repo with multiple implementations, that can be found in here: https://github.com/PacktPublishing/Deep-Reinforcement-Learning-Hands-On
The root folder consits of:
- Different games, every game has a RL algorithm, models and graphs
- requirements.txt (probably not up to date)
- runTensorBoard {dir}, runs tensorboard on a choosen directory
- old_agents, implementations of weaker RL algorithms
When you want to try out trained model, you have to set the LEARN flag in the agent file to false. Different models are trained on different observations, so not every combination will work. But the models name indicates the settings for the parameters.
- dqn_v2, a simple dqn implementation that offers experience replay. This is currently the best algorithm in this repository.
- value_iteration
- cross_entropy
- others
- All variants of dqn up until rainbow dqn.
- AC2 and AC3
- PPO2
- more gamessss
- The environment is constantly being updated, thus saved models often are not usable anymore
- old_agents probably dont work because the folder structured changed often