Python Deep Learning - Second Edition

Types of RL algorithms

We can use different classification of RL algorithms, depending on several factors. In this section, we'll outline some of the algorithms.

First, we'll divide RL algorithms based on the nature of the value function. We can identify two main types:

• Tabular solutions: The number of possible states and actions is small enough that we can represent the value function as a table (array) and the agent is fully familiar with the environment. One such scenario is the maze example, where the whole maze is stored in a table and the maze itself is not too big. With tabular solutions, we can often find the true optimal value function and optimal policy.
• Approximate solutions: The state and action spaces could be arbitrarily large. ...