Hands-On Reinforcement Learning with Python

Policy iteration

Unlike value iteration, in policy iteration we start with the random policy, then we find the value function of that policy; if the value function is not optimal then we find the new improved policy. We repeat this process until we find the optimal policy.

There are two steps in policy iteration:

1. Policy evaluation: Evaluating the value function of a randomly estimated policy.
2. Policy improvement: Upon evaluating the value function, if it is not optimal, we find a new improved policy:

The steps involved in the policy iteration are as follows:

1. First, we initialize some random policy
2. Then we find the value function for that ...