Hands-On Intelligent Agents with OpenAI Gym

Customizing the Atari Gym environment

Sometimes, we may want to change the way the observations are sent back by the environment or change the scale of the rewards so that our agents can learn better or filter out some information before the agent receives them or change the way the environment is rendered on the screen. So far, we have been developing and customizing our agent to make it act well in the environment. Wouldn't it be nice to have some flexibility around how and what the environment sends back to the agent so that we can customize how the agent learns to act? Fortunately, the Gym library makes it easy to extend or customize the information sent by the environment with the help of Gym environment wrappers. The wrapper interface ...