OfflineRL-Kit

• Github repo: OfflineRL-Kit.

OfflineRL-Kit is an offline reinforcement learning library based on pure PyTorch. This library has some features which are friendly and convenient for researchers, including:

• Elegant framework, the code structure is very clear and easy to use
• State-of-the-art offline RL algorithms, including model-free and model-based approaches
• High scalability, you can build your new algorithm with few lines of code based on the components in our library
• Support parallel tuning, very convenient for researchers
• Clear and powerful log system, easy to manage experiments

Supported algorithms

• Model-free
  • Conservative Q-Learning (CQL)
  • TD3+BC
  • Implicit Q-Learning (IQL)
  • Ensemble-Diversified Actor Critic (EDAC)
  • Mildly Conservative Q-Learning (MCQ)
• Model-based
  • Model-based Offline Policy Optimization (MOPO)
  • Conservative Offline Model-Based Policy Optimization (COMBO)
  • Robust Adversarial Model-Based Offline Reinforcement Learning (RAMBO)
  • Model-Bellman Inconsistency Penalized Offline Policy Optimization (MOBILE)