Reinforcement learning

Here is a short summary of a project I recently completed during my study of reinforcement learning algorithms.

I believe, if you really want to understand some algorithm, you should write it yourself.

That is why I implemented several algorithms from scratch and ensured they work as intended:

• Covariance Matrix Adaptation for Evolution Strategy (CMA-ES).
• Deep Q-Network (DQN), with options: Duelling-DQN, …
• Asynchronous Advantage Actor-Critic (A3C)
• Advantage Actor-Critic (A2C), the synchronized version of A3C, yet with multiple worker threads.
• Proximal Policy Optimization (PPO)

PPO demonstrated the best results.

For more details:

View Project on Github