Hands-On Q-Learning with Python

Algorithms

The algorithms environments let us train an agent to learn to imitate algorithmic tasks, such as copying characters and adding numbers together. As in any other RL task, the agent is not explicitly told what the task is; it learns through trial and error how to perform the task and when it has performed a correct action:

The copy task, shown in the preceding screenshot, involves copying the characters from the observation tape to the output tape.

Regardless of which RL model we choose to solve this or any other task, the model has to learn what the task actually is (Add the following symbols as they appear on the observation tape ...