gym-cards

This is a collection of gym environments for training reinforcement algorithms to play simple card games. This environment runs inside the OpenAI Gym.

Available environments

gym_cards:Wizards-v0

A simple card game with thew following features:

multiple agents (parameter players)
simulated agents play legal cards in order they were dealed
cards of multiple suits and different values (parameters suits and max_card)
players have to play the same suit as the first player or lose the game (-100 points)
the highest card wins a round, each round awards one point to the winner

Running environments

Install this package using the following commands.

git clone https://github.com/stoman/gym-cards.git
pip3 install -e gym-cards

All environments can now be used as the default Gym environments.

In particular, you can run the baselines after installing them:

export OPENAI_LOG_FORMAT='stdout,log,csv,tensorboard'
python3 -m baselines.run --env=gym_cards:Wizards-v0 --log_path=~/logs/wizards --save_path=~/models/wizards_ppo2 --num_timesteps=1e7

To monitor progress run tensorboard:

tensorboard --logdir ~/logs/wizards