Project · Python · Reinforcement learning · Gym environment · MLflow
Fantasy RL
A reinforcement-learning environment where AI agents learn to manage fantasy-football teams.
Fantasy RL explores how reinforcement-learning agents can optimize fantasy-football team management through player trading, team composition, budget management, and strategic decision-making.
The environment
The project simulates a fantasy-football league with a custom environment. Agents can buy and sell players, manage a limited budget, finish weekly transfers, and receive rewards based on team performance.
The state includes team metrics, market data, budget information, and normalized player indicators. This creates a controlled environment for experimenting with different strategies and reward systems.
Algorithms and experiments
The repository includes implementations of DQN, PPO, REINFORCE, and SARSA. Training runs produce reward plots, model checkpoints, and statistics, while MLflow is used to track experiments and compare results.
Relationship with Fantasy API
Fantasy RL uses Fantasy API as its data layer for collecting and organizing player, team, match, and performance data. Together, the projects explore the connection between sports analytics, data engineering, and machine learning.