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 SmoothL1LossToo many episodes / Stuck for long (low epsilon_min)Increase epsilon_min (0.01) - should reduce episodesGot unstuck after rare explorationalmost beat level Episode: 4748 03-10-2020 Increased epsilon_min (0.09) used less episodesbeat level after more exploration Episode: 2603 Episode: 3545
