A PyTorch reinforcement learning library implementing DQN, DDPG, A2C, PPO, SAC, MADDPG, A3C, APEX, and IMPALA.
Machin is a reinforcement learning library designed for PyTorch that implements a wide range of modern RL algorithms including DQN, DDPG, A2C, PPO, SAC, MADDPG, A3C, APEX, and IMPALA. It provides a simple and clear implementation of these algorithms to solve complex decision-making problems in custom environments.
Researchers and developers working on reinforcement learning projects who prefer PyTorch and need a library with readable, reusable implementations of advanced RL algorithms.
Developers choose Machin for its minimal abstractions, detailed documentation, and straightforward approach similar to PyTorch, making it easier to adapt algorithms to custom environments compared to more complex frameworks.
Reinforcement learning library(framework) designed for PyTorch, implements DQN, DDPG, A2C, PPO, SAC, MADDPG, A3C, APEX, IMPALA ...
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Integrates with PyTorch Lightning to generate and launch experiment configs automatically, reducing manual setup for standard RL workflows as shown in the auto module.
Uses minimal abstractions for clarity, making it easy to understand and modify code, aligning with its PyTorch-like philosophy emphasized in the documentation.
Encapsulates algorithms in classes for direct import, and leverages PyTorch's RPC API for building custom distributed training programs, as highlighted in the features.
Supports a wide range of modern RL methods from DQN to IMPALA, including single-agent, multi-agent, and parallel algorithms, detailed in the supported algorithms list.
Distributed algorithms and some functions are not fully supported on Windows or macOS, requiring platform-specific test scripts and potentially limiting cross-platform deployment.
Tests are 'weakly' reproducible and may not match original paper environments exactly, which could affect research validation and comparison efforts.
Missing key algorithms like QMIX and model-based methods listed in the roadmap, limiting immediate use for advanced multi-agent or model-based RL projects.