This is my implementation of "Rainbow is all you need".

Performance of Rainbow (from src/08.ipynb)

Index

• 01: Vanilla DQN
• 02: Double DQN
• 03: Prioritized Experience Replay
• 04: Dueling Network
• 05: Noisy Network
• 06: Categorical DQN
• 07: N-step Learning
• 08: Rainbow DQN (combination of all improvements)

Running

• pip install from requirements.txt
• Run the notebooks.

TODO

• 04: add visualization graphs for Advantage (max, min, mean) and Value
  • maybe with a fill between?