gym-yahtzee

Yahtzee game using OpenAI Gym meant to be used specifically for Reinforcement Learning. The rules are a loose interpretation of the free choice Joker rule, where an extra yahtzee cannot be substituted for a straight, where upper section usage isn't enforced for extra yahtzees. The maximum score is 1505, as opposed to 1375 using traditional forced choice Joker rules. In practice this doesn't affect gameplay significantly.

Example

To run a single game try the code below. As Gym doesn't support changing the action_space during a run, gym-yahtzee provides the function env.sample_action() which only samples from valid actions, e.g. no dice reroll after three rolls. Calling env.action_space.sample() also works, but will take longer to complete. Calling an invalid action results in a reward of zero and replaying the round.

from gym_yahtzee.envs.yahtzee_env import YahtzeeSingleEnv
import gym

env: YahtzeeSingleEnv = gym.make('yahtzee-single-v0')
env.reset()
for t in range(1000):
    env.render()
    action = env.sample_action()
    observation, reward, done, info = env.step(action)
    if done:
        print("Episode finished after {} timesteps".format(t+1))
        break

Developers guide

Pipenv is recommended for setting up a development environment. Prior to installing pipenv, creating a .env file with the following contents is recommended:

PYTHONPATH=.

To install pipenv and the required dependencies run the following commands:

pip install pipenv
pipenv install -r requirements.txt
pipenv shell

Updating dependencies

requirements.txt is dynamically generated using pip-compile. To regenerate the requirements.txtfile run the following command:

pip-compile -U --output-file requirements.txt setup.py requirements-dev.in