Reinforced-lib: Reinforcement learning library

Introducing Reinforced-lib: a lightweight Python library for the rapid development of reinforcement-learning (RL) solutions. It is open-source, prioritizes ease of use, provides comprehensive documentation, and offers both deep reinforcement learning (DRL) and classic non-neural agents. Built on JAX, it facilitates exporting trained models to embedded devices, and makes it great for research and prototyping with RL algorithms. Access to JAX's just-in-time (JIT) compilation ensures high-performance results.

Installation

You can install the latest version of Reinforced-lib from PyPI:

pip install reinforced-lib

To have easy access to the example files you can clone the source code from our repository, and then install it locally with pip:

git clone [email protected]:m-wojnar/reinforced-lib.git
cd reinforced-lib
pip install -e .

In the spirit of making Reinforced-lib a lightweight solution, we include only the necessary dependencies in the base requirements. To fully benefit from Reinforced-lib's conveniences, such as TF Lite export, install with the "full" suffix:

pip install ".[full]"

Key components

Reinforced-lib facilitates seamless interaction between RL agents and the environment. Here are the key components within of the library, represented in the API as different modules.

• RLib - The core module which provides a simple and intuitive interface to manage agents, use extensions, and configure the logging system. Even if you're not an RL expert, RLib makes it easy to implement the agent-environment interaction loop.

• Agents - Choose from a variety of RL agents available in the Agents module. These agents are designed to be versatile and work with any environment. If needed, you can even create your own agents using our documented recipes.

• Extensions - Enhance agent observations with domain-specific knowledge by adding a suitable extension from the Extensions module. This module enables seamless agent switching and parameter tuning without extensive reconfiguration.

• Loggers - This module allows you to monitor agent-environment interactions. Customize and adapt logging to your specific needs, capturing training metrics, internal agent state, or environment observations. The library includes various loggers for creating plots and output files, simplifying visualization and data processing.

The figure below provides a visual representation of Reinforced-lib and the data-flow between its modules.

JAX Backend

Our library is built on top of JAX, a high-performance numerical computing library. JAX makes it easy to implement RL algorithms efficiently. It provides powerful transformations, including JIT compilation, automatic differentiation, vectorization, and parallelization. Our library is fully compatible with DeepMind's JAX ecosystem, granting access to state-of-the-art RL models and helper libraries. JIT compilation significantly accelerates execution and ensures portability across different architectures (CPUs, GPUs, TPUs) without requiring code modifications. JAX offers another benefit through its robust pseudorandom number generator system, employed in our library to guarantee result reproducibility. This critical aspect of scientific research is frequently underestimated but remains highly significant.

Edge Device Export

Reinforced-lib is designed to work seamlessly on wireless, low-powered devices, where resources are limited. It's the perfect solution for energy-constrained environments that may struggle with other ML frameworks. You can export your trained models to TensorFlow Lite with ease, reducing runtime overhead and optimizing performance. This means you can deploy RL agents on resource-limited devices efficiently.

Example code

Experience the simplicity of our library and witness the fundamental agent-environment interaction loop with our straightforward example. This code can by used to train a deep Q-learning agent on the CartPole-v1 environment effortlessly using Reinforced-lib.

import gymnasium as gym
import optax
from chex import Array
from flax import linen as nn

from reinforced_lib import RLib
from reinforced_lib.agents.deep import DQN
from reinforced_lib.exts import Gymnasium


class QNetwork(nn.Module):
    @nn.compact
    def __call__(self, x: Array) -> Array:
        x = nn.Dense(256)(x)
        x = nn.relu(x)
        return nn.Dense(2)(x)


if __name__ == '__main__':
    rl = RLib(
        agent_type=DQN,
        agent_params={
            'q_network': QNetwork(),
            'optimizer': optax.rmsprop(3e-4, decay=0.95, eps=1e-2),
        },
        ext_type=Gymnasium,
        ext_params={'env_id': 'CartPole-v1'}
    )

    for epoch in range(300):
        env = gym.make('CartPole-v1', render_mode='human')

        _, _ = env.reset()
        action = env.action_space.sample()
        terminal = False

        while not terminal:
            env_state = env.step(action.item())
            action = rl.sample(*env_state)
            terminal = env_state[2] or env_state[3]

Citing Reinforced-lib

To cite this repository, please use the following BibTeX entry for the Reinforced-lib paper:

@article{reinforcedlib2022,
  author = {Maksymilian Wojnar and Szymon Szott and Krzysztof Rusek and Wojciech Ciezobka},
  title = {{R}einforced-lib: {R}apid prototyping of reinforcement learning solutions},
  journal = {SoftwareX},
  volume = {26},
  pages = {101706},
  year = {2024},
  issn = {2352-7110},
  doi = {https://doi.org/10.1016/j.softx.2024.101706},
  url = {https://www.sciencedirect.com/science/article/pii/S2352711024000773}
}