Simulator and Agents for the game of Love Letter.
Love Letter is a turn based card game where a player wins by eliminating the other players.
Numbers match the win rate of the agent on the bottom row against the opponent agent in the left column. Darker colors are better - the best agent has the column of darkest colors. Note these are all based on four player games, so a random agent will win 25% of the time against three other random agents.
This agent chooses from valid actions randomly every time.
Based on a system covered in the paper Asynchronous Methods for Deep Reinforcement Learning, this system involves multiple neural networks updating based on experienced episodes.
Run with python play.py.
Requirements are contained in requirements.txt. Install with pip install -r requirements.txt.
Run with make test, python loveletter/setup.py test, or python -m unittest discover.