Coursework for Deep Reinforcement Learning

Authors: Alex Collins and Khalil Adib

Added 25 May

For the advanced part, code is all in the main folder. Python files:

• Game.py - sets up the environment, and has methods to aid the snake's perception of it
• Agent.py - controls the behaviour of the sname
• Model.py - the neural nets used by the agent (DQN or Dueling)
• ExperienceReplayBuffer.py - class to control the replay buffer
• helper.py - visualisation functions and more importantly the training loop

Analysis is presented in the notebooks. Examiners can ignore everything except the snappily named Q-Learning_Comparisons-v2.ipynb. This has all training for all models, all evaluation for all models/approaches and the best explanations. Everything else can by ignored.