fordb / gym-cribbage

RL for the cribbage card game

gym-cribbage

An environment for playing Cribbage in OpenAI gym (2-4 players).

setup

The gym environment can be installed using python setup.py install.

development

Run pip install -e ..

usage

The environment can be imported after install via:

from gym_cribbage.envs.cribbage_env import CribbageEnv
env = CribbageEnv(n_players=2, verbose=False)

Where n_players can be between 2-4 and verbose can be set to True for (copius) debug information.

The environment can be initalized via:

state, reward, done, debug = env.reset()

Which starts a new game. The environment can be interacted with as a standard openAI gym environment, I.e., env.step(), env.render() and env.reset().

The state returned by the environment contains a State() object with the following fields:

• State.hand -- the current player's playable cards. Non-playable cards are never returned in the hand.
• State.hand_id -- the ID of the current player.
• State.reward_id -- the ID of the player who recieves the reward returned by the environment.
• State.phase -- the current phase of the game. 0 == 'the Deal', 1 == 'the Play', 2 = 'the Show'.
• State.player_score -- the current score (out of 121) of the current player.
• State.opponent_score -- a list of all opponent scores (out of 121).

The cribbage environment cycles through hands an accumulates scores until one player reaches 121, whereby the environment immediately returns done==True and the game is over. A new game can be started via state, reward, done, debug = env.reset().

Actions taken during steps have different meaning during each phase of the game.

• During 'the Deal' (State.phase == 0), cards submitted via env.step(Card) are discarded to the crib. Cards must be taken from the player's hand State.hand[i].
• During 'the Play' (State.phase == 1), cards submitted via env.step(Card) are played to the table to earn points. Again, the must be taken from the player's hand State.hand[i].
• During 'the Show' (State.phase == 2), nothing submitted is used by the environment. Therefore it is convienient to pass a 'dummy card' to the environment state, reward, done, debug = env.step(Card(99, SUITS[0])), which can be easily done if one imports some utility classes from cribbage_env: from gym_cribbage.envs.cribbage_env import Card, SUITS. The purpose of these steps is to return the appropriate points to each agent for Show (in sequence, following the rules of Cribbage).

Rules

https://en.wikipedia.org/wiki/Cribbage