Blackjack custom Gymnasium environment game

Project focuses on a Blackjack game implementation using the Python library Gymnasium. The game environment can be used for reinforcement learning projects.

Currently, there are two versions of the Blackjack environment:

• Blackjack4game-v0 - basic implementation with discrete action space
• Blackjack4game-v1 - implementation featuring betting at the beggining of a game - where an agent can decide what percentage of money to bet (continuous action space)

Blackjack rules

• in Polish
• in English

Note: The current implementation does not support SPLIT moves. Besides that, most other standard rules apply. There are still areas for improvement - some of them are marked with #TODO comment in the code.

Requirements

• Python
  configuration file (pyproject.toml)

Usage

1. Clone the repository:

   git clone https://github.com/jakseluz/blackjack_gym.git
   cd blackjack_gym

2. Install project and its dependencies with:

   pip install -e .

   • (in the project root directory; see the point above)

3. Use the environment in a selected version:

   import gymnasium as gym
   import blackjack_env
   
   # version 0 (see README description above)
   env = gym.make("Blackjack4game-v0", render_mode="terminal") # render_mode="human" if you want to use GUI (or None if neither)
   
   # version 1
   # env = gym.make("Blackjack4game-v1", render_mode="terminal")
   
   # use it like other gymnasium environments;

4. To see how to use the gymnasium environment, see: gymnasium documentation.

5. Check ./notebooks/ directory for the tutorial/ information:

   • ./notebooks/main.ipynb file - Blackjack4game-v0 tutorial ./notebooks/bet_example.ipynb file - additional info about Blackjack4game-v1 version

6. More detailed analysis available (in Polish) in

   • ./notebooks/report.ipynb
   • ./notebooks/report_continuous_env.ipynb

7. Result files used in learning etc. are located in ./results/:

   • q_learning files
   • q_learning plots
   • PPO files - next to the above folders in seperate directories inside the ./results/ catalogue

Code

1. The environment directory:
   • Blackjack4game-v0
   • Blackjack4game-v1
2. Agent learning utilities.
3. Wrappers used by the v1 version.

Authors

• Jakub Łabuz (jakseluz) - Blackjack
• Maciej Maj (mmaj1) - most of the tests and reports

Used resources:

• Playing Card Deck
• Gymnasium Environment Template

---

For questions or contributions, please open an issue or submit a pull request!