Название: A Primer to the 42 Most commonly used Machine Learning Algorithms (With Code Samples)
Автор: Murat Durmus
Издательство: Leanpub
Год: 2023-01-28
Страниц: 192
Язык: английский
Формат: pdf (true), epub, mobi
Размер: 10.2 MB

This book introduces you to the 42 most commonly used machine learning algorithms in an understandable way.

Machine Learning (ML) refers to the development of AI systems that can perform tasks due to a “learning process” based on data. This is in contrast to approaches and methods in symbolic AI and traditional software development, which are based on embedding explicit rules and logical statements in the code. ML is at the heart of recent advances in statistical AI and the methodology behind technological achievements such as computer programs that outperform humans in tasks ranging from medical diagnosis to complex games. The recent surge of interest in AI is largely due to the achievements made possible by ML. As the term “statistical AI” suggests, ML draws on statistics and probability theory concepts. Many forms of ML go beyond traditional statistical methods, which is why we often think of ML as an exciting new field. However, despite the hype surrounding this technological development, the line between ML and statistics is blurred. There are contexts in which ML is best viewed as a continuum with traditional statistical methods rather than a clearly defined separate field. Regardless of the definitional boundaries, ML is often used for the same analytical tasks that conventional statistical methods have been used for in the past.

ML Approaches: ML is a very active area of research that encompasses a broad and ever-evolving range of methods. Three primary approaches can be distinguished at a high level: supervised learning, unsupervised learning, and reinforcement learning.

Supervised Learning: In supervised learning, the task of the ML algorithm is to infer the value of a predefined target variable (or output variable) based on known values of feature variables (or input variables). The presence of labeled data (i.e., data with known values for the target in question) is a prerequisite for supervised learning. The learning process consists of developing a model of the relationship between feature and target variables based on labeled training data. This process is also referred to as “model training.” After a successful training phase (which is confirmed by a testing phase also based on labeled data), the resulting model can be applied to unlabeled data to infer the most likely value of the target variable. This is referred to as the inference phase.

Unsupervised learning involves identifying patterns and relationships in data without a predefined relationship of interest. Unlike supervised learning, this approach does not rely on labeled training data. Therefore, unsupervised learning can be more exploratory, although the results are not necessarily less meaningful.

Reinforcement learning is based on the concept of an “agent” exploring an environment. The agent’s task is to determine an optimal action or sequence of steps (the goal of interest) in response to its environment. The learning process does not rely on examples of “correct responses.” Instead, it depends on a reward function that provides feedback on the actions taken. The agent strives to maximize its reward and thus improve its performance through an iterative process of trial and error.

The following algorithms are covered in this book: