RL as a Markov decision process

A Markov decision process (MDP) is a mathematical framework for modeling decisions. We can use it to describe the RL problem. We'll assume that we work with a full knowledge of the environment. An MDP provides a formal definition of the properties we defined in the previous section (and adds some new ones):

• is the finite set of all possible environment states, and s_t is the state at time t.
• is the set of all possible actions, and a_t is the action at time t.
• is the dynamics of the environment (also known as transition probabilities matrix). It defines the conditional probability of transitioning to a new state, s', given the existing state, s, and an action, a (for all states and actions):

We have transition probabilities between the states, because MDP is stochastic (it includes randomness). These probabilities represent the...