kuzongaenv 0.3.3

Gymnasium environment for the game Kuzonga

KuzongaEnv

A custom Gymnasium-compatible environment for the Kuzonga game.

Environment Details

Action Space

The environment uses a dictionary action space with three components:

Key	Value	Description
v	1 or 0 (or True and False, respectively)	Whether to attempt division (1) or change a digit (0).
g	0–9	If v=1, the divisor; if v=0, the new digit to set at r.
r	0…digits-1 or None	Rindex (Right-to-left or reverse index) of the digit to overwrite (if v=1, it should be None).

Example:

action = {"v": 1, "g": 3, "r": None}  # attempt division by 3
action = {"v": 0, "g": 7, "r": 1}  # set the second digit (from the right) to 7

Observation Space

The environment uses a dictionary observation space with the following keys:

Key	Type	Description
s	np.int8 array (digits,)	The original number as an array of digits.
d	np.int8 array (digits,)	The current number as an array of digits.
a	np.int64 array (digits*10,)	Binary mask of which digits can be set at each position. Flattened from shape (digits, 10).
p	np.int64 array (num_players*3,)	Each player’s [i, c, m], where i is the ID, c is the score and m tells if it is the player's turn (m=1) or not (m=0). Flattened array of all players. It has one player by default.
t	int	ID of the player whose turn it is.

Example:

obs, info = env.reset()
print(obs["s"])   # [1, 7]
print(obs["d"])   # [4, 7]
print(obs["a"])   # array([1,1,0,...])
print(obs["p"])   # array([0,0,1,1,0,0])  # two players
print(obs["t"])   # 0

Quick Notes

The a mask ensures illegal moves (e.g., setting a leading zero or creating number 0/1) are prevented.

Rewards and penalties are automatically updated in the environment during step().

The environment fully supports multiple players, and tracks turns via t and m.

The options parameter in reset() allows resetting the environment to a specific given state/obs, by setting it with the format: options = {'obs': <state/obs dict>}

The info variable value from methods reset() and step() has the fields:

• obs_decoded, which has the plain values of the resulting state/observation. So it can be retrieved as info['obs_decoded'].
• all_actions, which is a list of all the actions in the current state. So it can be retrieved as info['all_actions'].
• concluded, which is True if the state is a finished state (winning game conditions have been met), and False otherwise. So it can be retrieved as info['concluded'].

Usage Example

import gymnasium as gym
import kuzongaenv

env = gym.make("Kuzonga-v0")
obs, info = env.reset()
action = env.action_space.sample()
obs, reward, terminated, truncated, info = env.step(action)

print(f"Observation: {obs}")
print(f"Reward: {reward}, Terminated: {terminated}")

Installation

pip install -e .

Cite This Project

If you use Kuzonga in your research, projects, or publications, please cite it as:

Jacinto Jeje Matamba Quimua (2025). KuzongaEnv: Gym Environment for Reinforcement Learning Experiments. GitHub repository: https://github.com/jaci-hub/kuzongaenv

BibTeX

@misc{kuzongaenv2025,
  author       = {Jacinto Jeje Matamba Quimua},
  title        = {KuzongaEnv: Gym Environment for Reinforcement Learning Experiments},
  year         = 2025,
  howpublished = {\url{https://github.com/jaci-hub/kuzongaenv}},
}

Play Kuzonga Online

Kuzonga game