Keras Reinforcement Learning Projects

SARSA

As we anticipated in Chapter 1, Overview of Keras Reinforcement Learning, the State-Action-Reward-State-Action (SARSA) algorithm implements an on-policy TDs method, in which the update of the action value function (Q) is performed based on the results of the transition from the state s = s (t) to the state s' = s (t + 1) by the action a (t), taken on the basis of a selected policy: π (s, a).

There are policies that always choose the action that provides the maximum reward, and non-deterministic policies (ε-greedy, ε-soft, and softmax), which ensure an element of exploration in the learning phase.

In SARSA, it is necessary to estimate the action-value function q(s, a), because the total value of a state v(s) (value function) is not sufficient ...