jaxagents 0.1.4

Implementation of Reinforcement Learning agents in JAX

Jaxagents

Jaxagents is a Python implementation of Reinforcement Learning agents built upon JAX.

Content

So far, the project includes the following agents:

• Q-learning:
  • Deep Q Networks (DQN)
  • Double Deep Q Networks (DDQN)
  • Categorical Deep Q Networks (often known as C51)
  • Quantile Regression Deep Q Networks (QRDQN)
• Policy gradient:
  • REINFORCE
  • PPO with clipping

Background

Research and development in Reinforcement Learning can be computationally cumbersome. Utilizing JAX's high computational performance, Jaxagents provides a framework for applying and developing Reinforcement Learning agents that offers benefits in:

• computational speed
• easy control of random number generation
• hyperparameter optimization (via parallelized calculations)