Reinforcement Learning Algorithms with Python

The policy

In the case that the actions are discrete and limited in number, the most common approach is to create a parameterized policy that produces a numerical value for each action.

Note that, differently from the Deep Q-Network algorithm, here, the output values of the policy aren't the Q(s,a) action values.

Then, each output value is converted to a probability. This operation is performed with the softmax function, which is given as follows:

The softmax values are normalized to have a sum of one, so as to produce a probability distribution where each value corresponds to the probability of selecting a given action.

The next two plots ...