Investigated Libraries

• OpenAI baselines in python and TensorFlow includes many state of the art algorithms.
• RLLib in python, for TensorFlow and PyTorch, includes tool for hyperparameter tuning, and very importantly support for multi-agent systems.
• Nervana’s Coach in python and TensorFlow and optimised for Intel processors, also with many relevant algorithms.
• TensorForce in python and TensorFlow is an ambitious framework that also includes many relevant algorithms.
• RLLab(not active) implements in python very relevant algorithms like Trust Region Policy Optimization (TRPO) and Deep Deterministc Policy Gradient (DDPG) algorithms, Cross-Entropy Method (CEM), Covariance Matrix Adaptation - Evolutionary Strategy (CMA-ES)… and some others.
• Keras-RL(not active) implements in python Deep Q-learning (DQN), Double DQN (which removes the bias from max operator in Q-learning), DDPG, Continuous DQN (CDQN or NAF) and CEM.
• tf_agents

Reference

• https://www.quora.com/What-are-some-of-the-good-Reinforcement-Learning-libraries