Agent Stubs

The coax package doesn’t include full-fledged agents. The reason is that it is designed for RL practitioners who want to create their own agents using a modular API.

For those who would like to have a starting point for creating pre-existing agents such as DQN or PPO we have collected simple Agent Stubs here. These stubs are partially complete python scripts that show the general structure of an agent.

Value-Based Agents

• SARSA
• Q-Learning
• Soft Q-Learning
• Deep Q-Network (DQN)
• Prioritized Experience Replay (PER)
• Implicit Quantile Network (IQN)

Policy-Based Agents

• REINFORCE
• Advantage Actor-Critic (A2C)
• Proximal Policy Optimization (PPO)
• Deep Deterministic Policy Gradients (DDPG)
• Twin-Delayed DDPG (TD3)