Hackathon Machine Intelligence HS22
Run dqn: python main.py --algorithm dqn --n_episodes 1000
Task:
Your task is to train a deep reinforcement learning agent in the OpenAI-Gym "LunarLander-v2" environment.
(Notice that development of Gym has been moved to Gymnasium. You can use either the former Gym library, or the new one in Gymnasium)
The goal is to come up with your best performing agent and see how it compares to the other Hackathon teams!
-
try different algorithms: DQN, AC, DDPG, ... (you are strongly encouraged to develop your own implementations, not just use examples found on github or provided ones from "OpenAI Baselines". But of course, you can get use them as inspiration, starting points or benchmarks.)
-
You should study and try to improve the agent in detail (e.g. vary hyperparameters, network architecture etc.)
-
You should measure and evaluate performance (Reward vs training episodes etc.), do numerical comparisons, charts, diagrams etc.
-
If you have time, compare with the continuous action space variant of the environment ("LunarLanderContinuous-v2")
Formalities:
The hackathon will be developed in teams of two people.
Timeframe: Wednesday, 11.01.2023, 09:00 – Friday, 13.01.2022, 15:30
Final team presentations and Q&A: Friday, 13..01.2023, 15:00-16:30
Grading:
Grading will be based on the team presentations (20 minutes per team, plus 10 minutes Q&A). Both team members are expected to present The presentation should contain: Methodology, Results, Code Overview, & brief demo. Part of the demo can be presented during the Q&A section.
Links:
LunarLander-V2 in Gym: https://www.gymlibrary.dev/environments/box2d/lunar_lander/ LunarLander-V2 in Gymnasium: https://gymnasium.farama.org/environments/box2d/lunar_lander/