Markov Decision Process
MDP is an extension of the Markov chain. It provides a mathematical framework for modeling decision-making situations. Almost all Reinforcement Learning problems can be modeled as MDP.
MDP is represented by five important elements:
- A set of states
the agent can actually be in. - A set of actions
that can be performed by an agent, for moving from one state to another. - A transition probability (
), which is the probability of moving from one state 
to another 
state by performing some action 
. - A reward probability (
), which is the probability of a reward acquired by the agent for moving from one state 
to another state 
by performing some action 
. - A discount factor (
), which controls the importance of immediate and future rewards. We will discuss this in detail in the upcoming sections.