divide21env 0.2.4

Gymnasium environment for the game Divide21

Divide21Env

A custom Gymnasium-compatible environment for the Divide21 game.

Environment Details

Action Space

The environment uses a dictionary action space with three components:

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

Example:

action = {"division": 1, "digit": 3, "rindex": None}  # attempt division by 3
action = {"division": 0, "digit": 7, "rindex": 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
static_number	np.int8 array (digits,)	The original number as an array of digits.
dynamic_number	np.int8 array (digits,)	The current number as an array of digits.
available_digits_per_rindex	np.int64 array (digits*10,)	Binary mask of which digits can be set at each position. Flattened from shape (digits, 10).
players	np.int64 array (num_players*3,)	Each player’s [id, score, is_current_turn]. Flattened array of all players. It has one player by default.
player_turn	int	ID of the player whose turn it is.

Example:

obs, info = env.reset()
print(obs["static_number"])             # [1, 7]
print(obs["dynamic_number"])            # [4, 7]
print(obs["available_digits_per_rindex"])  # array([1,1,0,...])
print(obs["players"])                   # array([0,0,1,1,0,0])  # two players
print(obs["player_turn"])               # 0

Quick Notes

The available_digits_per_rindex 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 player_turn and is_current_turn.

The manual_obs parameter in reset() allows resetting the environment to a specific given state/obs.

Usage Example

import gymnasium as gym
import divide21env

env = gym.make("Divide21-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 Divide21 in your research, projects, or publications, please cite it as:

Jacinto Jeje Matamba Quimua (2025). Divide21Env: Gym Environment for Reinforcement Learning Experiments. GitHub repository: https://github.com/jaci-hub/divide21Env

BibTeX

@misc{divide21env2025,
  author       = {Jacinto Jeje Matamba Quimua},
  title        = {Divide21Env: Gym Environment for Reinforcement Learning Experiments},
  year         = 2025,
  howpublished = {\url{https://github.com/jaci-hub/divide21Env}},
}

Play Divide21 Online

Divide21 game