What is Reinforcement Learning?
Reinforcement Learning(RL) is the task of learning through trail and error without the need of human supervision.
Example: Ping-Pong playing agent, the agent tries to learn through a simulator and trains in such a way to successfully play ping-pong.
Components of Reinforcement Learning
Two core components of RL system are the agent and the environment.
Agent is the decision maker and problem solver the environment is the representative of that problem.
Agent is the decision maker in RL, it works by three steps process.
a. Agents have an interaction component which means a way to gather data from environment.
b. They evaluate their current performance.
c. They improve something in their inner workings after observation and learning.
Environment is the set of variables related to the problem.
State Space is the combination of all the variables that environment consists of.
Agents have the ability to observe from the environment, these set of variables that an agent perceives is called observation.
After every step the agent has options to choose some actions, the set of all actions from which agents can choose from is called action space.
Agents can influence the environment through its actions, the environment can change states due to this behavior of agent, the function is called transition function.
After this transition the environment gives a new observation and may also give a reward signal to agent based on its behavior, this function is called reward function.
References:
Grokking Deep Reinforcement Learning by Miguel Morales.
