sparse-Q-learning

A multi-agent system Q-learning with sparse cooperation

Right now, the best joint action is computed with a naive method and not with the elimination algorithm : we just compute the payoff for every combination of the action and we take the best

Launch

1. learn mode

Let the agents learn a policy during n episodes

python3 main.py learn [directory] -e episode -g grid -v

• directory : directory to store the rules file
• episode : number of episode
• grid : grid size of the prey-predators game

2. play mode

Play the game with a learned policy

python3 main.py play [directory] -g grid

• directory : directory to store the rules file
• grid : grid size of the prey-predators game

3. test mode

Test the performance of the learning

python3 main.py test [directory] -e episode -r run -g grid -v

• episode : number of episode
• run : number of run
• grid : grid size of the prey-predators game

Results

Game in the console

Class diagram