During the semester, the demonstrations and examples accompanying the RL lecture will be released on a weekly basis. They will stay up so they can be used to prepare the project and the exam.
- Recap & example demonstration (using notebook)
- Kahoot quiz
- Questions & discussion on the lecture & assignment
- Announcement of the new assignment:
- coding, most of the time implementing an important algorithm from the lecture
- one or two questions that check if the student actually understands the algorithm
- an open ended question on what the student observes for different exploration or learning algorithms or hyperparameters
- Mars Rover: moves left and right, could be extended to stay. W2, maybe also W3
- Cart Pole: a slight change of pace for W4 and 5, maybe even 6 so they can see a difference
- Continuous Mountain Car: a classic and a continuous env for a change. Maybe later with DL?
- A two player game so that we can do a christmas challenge
- Maybe something with MDP playgorund?
- policy & value iteration
- tabular SARSA & Q-Learning
- DQN from "scratch" in pytorch/tensorflow
- something using the OpenAI gym interface, so maybe their own env at some point? MDP playground
- at least epsilon greedy exploration, ideally more involved exploration