Lagom is a PyTorch-based infrastructure designed for rapid prototyping of reinforcement learning (RL) algorithms. It targets researchers and developers needing a balance between flexibility and usability, offering modular tools to quickly iterate on RL ideas without excessive boilerplate or loss of control.

How It Works

Lagom provides a modular framework built on PyTorch, abstracting common RL components like agents, environments, and training engines. It emphasizes a "not too much, not too little" philosophy, offering higher-level APIs for common tasks while retaining flexibility for custom implementations. The library includes built-in support for multiprocessing (master-worker) for parallel experiment execution and hyperparameter search capabilities (grid or random).

Quick Start & Requirements

• Install: Recommended via Miniconda (conda create -n lagom python=3.7), then pip install -r requirements.txt and pip install -e . for source installation.
• Prerequisites: Python 3.7, PyTorch.
• Documentation: http://lagom.readthedocs.io
• Examples: Available in the repository.

Highlighted Details

• Implements a collection of standard RL algorithms in baselines.
• Supports parallelization via a master-worker framework.
• Facilitates hyperparameter search (grid and random).
• Inspired by OpenAI Gym, Baselines, and Spinning Up.

Maintenance & Community

• Initial release in September 2018 by Zuo Xingdong.
• Last update noted in March 2019 (v0.0.3).

Licensing & Compatibility

• No license specified in the README.
• Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

The project's last update was in March 2019, indicating potential lack of recent maintenance or feature development. The absence of a specified license poses a significant blocker for adoption, especially in commercial or closed-source environments.