Final project in ELE680 Deep Neural Network at University of Stavanger (UiS)
In this project DQN is compared to Double-DQN in three games for the classic OpenAI Gym. The Double-DQN is a updated end slightly modified version of OpenAI's benchmark DQN, While the vanilla DQN is modified from that again.
The main focus of this scripts have been optimizing the hyperparameters using SigOpt. Due to limited compute the number of runs was not enough to give a definitive result. By running this project again it's recommended to increase number of episodes to above 10k and let SigOpt perform as many trials as possible.