Machine Learning: 2 Books in 1: Machine Learning for Beginners, Machine Learning Mathematics. An Introduction Guide to Understand Data Science Through the Business Application

Reinforcement Learning
Reinforcement learning is our third kind of machine learning. Like
unsupervised learning, no inputs are given. The reinforcement learning
model must discover on its own how to be the most effective. Then, the data
scientist makes suggests improving the model based on its ability to predict
an outcome, which is evaluated by looking at comparisons between
prediction and actual value. This is the most progressive type of machine
learning, and where much of the machine learning in the future will be
done.

---

Think of it like playing a game; over time, you learn by winning and losing.
You learn how to win by playing, and the more you become familiar with
the game, the better you understand the mechanics of winning. Over time,
the data scientist gives the model feedback as it collects and processes data.
It receives a reward signal so that it knows when it is doing a good job of
predicting some outcome. So ‘winning’ the game gives positive feedback to
the algorithm. This is the type of machine learning used in gaming, robots,
navigation, and self-driving cars.
Q Learning
In Q-learning, the model communicates with its environment to improve
itself. You begin by having a set of states. States are the things in the
environment which stand as obstacles and avenues in your environment.
Called “S.” In chess, it would be the way that all of your pieces could
move, as well as where all of your opponent’s pieces are. These are states.
The possible moves are called ‘A.' If you are a pawn, your possible moves
are one square forward. If you're a rook, your possible moves are in any
direction in a straight line. Q is the value of the model, which starts at 0. As
you play the game, Q goes up and down depending on its interactions with
the environment. With negative interactions, the score Q goes down. With
positive interactions, the score Q goes up. The algorithm learns how to
move so that it can optimize the number Q. In the beginning, it's random.
Over time, these random movements result in positive and negative effects
on Q, and the machine learns how each move will affect the score of Q. It
must play a lot of games to improve the way it plays over time. It's much
easier said than done to apply this process in real life.

---