PER-NAF

An implementation of the Normalized Advantage Function Reinforcement Learning Algorithm with Prioritized Experience Replay

Summary

• The original paper of this code is: https://arxiv.org/abs/1603.00748
• The code is mainly based on: https://github.com/carpedm20/NAF-tensorflow/
• Additionally I added the prioritized experience replay: https://arxiv.org/abs/1511.05952
• Using the OpenAI baseline implementation: https://github.com/openai/baselines/blob/master/baselines/deepq/replay_buffer.py

Thanks openAI and Kim!

Some Advices from experience in RL

• Normalize the state and action space as well as the reward is a good practice
• Visualise as much as possible to get an intuition about the method as possible bugs
• If it does not make sense it is a bug with very high probability

Coding makes happy 🙃