Hands-On Reinforcement Learning for Games

Actor-critic

AC methods use a combination of networks to predict the output of the value and policy functions, where our value function network resembles DQN and our policy function is defined using a PG method such as REINFORCE. Now, for the most part, this is as simple as it sounds; however, there are several details in the way in which we code these implementations that require some attention. We will, therefore, cover the details of this implementation as we review the code. Open Chapter_8_ActorCritic.py and follow the next exercise:

1. As this code follows the same pattern as previous examples, we will only need to cover a few sections in detail. The first section of importance is the new ActorCritic class at the top of the file and shown ...