HDS-LEE Hackathon „Hearts Gym” 2024

Hearts Gym Hackathon 2024

***The date is postponed. As soon as a new date has been set, you will be informed on the website.***

About the Hackathon

Learning reinforcement learning in a card game: Reinforcement learning is used to teach an agent optimal strategies (policy) to achieve maximum reward in a Markov decision process by using a simple card game: Hearts. In Hearts, four players play against each other and have to avoid winning hands that include hearts or the queen of spades. These players can be modelled as reinforcement learning agents. The Helmholtz AI team at FZJ created a multi-agent environment to train agents on playing the Hearts game, which also includes a client-server architecture to remotely evaluate local agents: https://github.com/HelmholtzAI-FZJ/hearts-gym. In this server architecture, the agents can battle against each other to evaluate which group has taught the best cards player. The overall aim is to do some Collaborative Coding within the HDS-LEE PhD student community as a networking event.   

Publication: https://arxiv.org/abs/2209.05466

Trainers

Jan Ebert and Stefan Kesselheim from Helmholtz AI team at FZJ

Prerequisites

Basic knowledge in machine learning (ML) and Python programming skills.

Install Python ≥ 3.6, < 3.9 and Git on your system.

Grouping

Size: 3–5 people

The groups should consists of at least one PhD with good knowledge in ML and one in Python. The group members can either find themselves or are drawn by lot according to these criteria.