rl-envs-forge 3.4.2

Environments for reinforcement learning

rl-envs-forge

Lightweight environments for reinforcement learning applications.

Table of Contents

• rl-envs-forge
  • Table of Contents
  • Installation
  • Environments
    • Labyrinth
      • Labyrinth rendered example
    • KArmedBandit
      • KArmedBandit rendered example
    • GridWorld
      • GridWorld rendered example
    • ACML
  • Usage
  • Tests
  • License
  • Contact & Support

Installation

pip install rl-envs-forge

Environments

Labyrinth

Labyrinth is a classic maze-solving environment, where the goal is to navigate from a start point to a target. The maze layout is randomly generated based on a set of parametrizable arguments.

📖 Detailed Documentation: Click here to read more about the Labyrinth environment

Labyrinth rendered example

KArmedBandit

KArmedBandit is a bandit environment, which returns a reward from a distribution associated with the chosen arm at each timestep. This implementation includes multiple distributions, and the possibility to shift the distribution parameters during sampling.

📖 Detailed Documentation: Click here to read more about the KArmedBandit environment

KArmedBandit rendered example

GridWorld

GridWorld is a customizable grid-based environment for reinforcement learning, featuring adjustable grid size, start and terminal states, walls, and special transitions. Each action taken by the agent results in a transition within the grid, adhering to the defined rules and probabilities.

📖 Detailed Documentation: Click here to read more about the GridWorld environment

GridWorld rendered example

ACML

The Adaptive Computation and Machine Learning (ACML) environments are toy environments proposed in Reinforcement Learning: An Introduction* (2nd ed.).

📖 Detailed Documentation: Click here to read more about the ACML environments

Usage

Example code on setting up and testing the Labyrinth environment.

Note, this code snippet produced the render visible in section Labyrinth

from time import sleep
from rl_envs_forge.envs.labyrinth.labyrinth import Labyrinth

env = Labyrinth(20, 20, seed=0)

done = False
quit_event = False
while not done and not quit_event:
    action = env.action_space.sample()  
    observation, reward, done, truncated, info = env.step(action)
    quit_event, _ = env.render()
    sleep(0.1)

Tests

Requirements: pytest and pytest-cov

Run the tests in the root folder with:

pytest tests

License

This project is licensed under the MIT License.

Contact & Support

For any queries or support, or if you would like to contribute to this project, reach out at [e-mail] or raise an issue on our GitHub repository.