RL-SuperMarioBros

Reinforcement learning algorithms implemented to play SuperMario. Will be used as a practical baseline to compare different algorithms & techniques.

Dueling Double DQN

Notes	GIFs
21-10-2020 epsilon_min (0.01) Very slow learning Score increased rarely post this episode	Episode: 4298

Dueling Double DQN with Prioritized Replay

Notes	GIFs
17-08-2020 Algorithm tested Agent didn't learn	NIL
02-10-2020 Switched to SmoothL1Loss Too many episodes / Stuck for long (low epsilon_min) Increase epsilon_min (0.01) - should reduce episodes Got unstuck after rare exploration almost beat level	Episode: 4748
03-10-2020 Increased epsilon_min (0.09) used less episodes beat level after more exploration	Episode: 2603 Episode: 3545