Overview of Book
While Bayesian networks have flourished in academia over the past three decades, their application for research has developed more slowly. One of the reasons has been the sheer difficulty of generating Bayesian networks for practical research and analytics use. For many years, researchers had to create their own software to utilize Bayesian networks. Needless to say, this made Bayesian networks inaccessible to the vast majority of scientists.
The launch of BayesiaLab 1.0 in 2002 was a major initiative by a newly-formed French company to address this challenge. The development team, lead by Dr. Lionel Jouffe and Dr. Paul Munteanu, designed BayesiaLab with research practitioners in mind—rather than fellow computer scientists. First and foremost, practitioner orientation is reflected in the graphical user interface of BayesiaLab, which allows researchers to work interactively with Bayesian networks in their native form using graphs, as opposed to working with computer code. At the time of writing, BayesiaLab is approaching its sixth major release and has developed into a software platform that provides a comprehensive “laboratory” environment for many research questions.
However, the point-and-click convenience of BayesiaLab does not relieve one of the duties of understanding the fundamentals of Bayesian networks for conducting sound research. With BayesiaLab making Bayesian networks accessible to a much broader audience than ever, demand for the corresponding training has grown tremendously. We recognized the need for a book that supports a self-guided exploration of this field. The objective of this book is to provide a practice-oriented introduction to both Bayesian networks and BayesiaLab.
This book reflects the inherently visual nature of Bayesian networks. Hundreds of illustrations and screenshots provide a tutorial-style explanation of BayesiaLab’s core functions. Particularly important steps are repeatedly shown in the context of different examples. The key objective is to provide the reader with step-by-step instructions for transitioning from Bayesian network theory to fully functional network implementations in BayesiaLab.
The fundamentals of the Bayesian network formalism are linked to numerous disciplines, including computer science, probability theory, information theory, logic, machine learning, and statistics. Also, in terms of applications, Bayesian networks can be utilized in virtually all disciplines. Hence, we meander across many fields of study with the examples presented in this book. Ultimately, we will show how all of them relate to the Bayesian network paradigm. At the same time, we present BayesiaLab as the technology platform, allowing the reader to move immediately from theory to practice. Our goal is to use practical examples for revealing the Bayesian network theory and simultaneously teaching the BayesiaLab technology.
The intention of the three short chapters in Part 1 of the book is to provide a basic familiarity with Bayesian networks and BayesiaLab, from where the reader should feel comfortable jumping into any of the subsequent chapters. For a more cursory observer of this field, Part 1 could serve as an executive summary.
The chapters in Part 2 are mostly self-contained tutorials, which can be studied out of sequence. However, beyond Chapter 8, we assume a certain degree of familiarity with BayesiaLab’s core functions.
- In Chapter 4, we discuss how to encode causal knowledge in a Bayesian network for subsequent probabilistic reasoning. In fact, this is the field in which Bayesian networks gained prominence in the 1980s in the context of building expert systems.
- Chapter 5 introduces data and information theory as a foundation for subsequent chapters. In this context, BayesiaLab’s data handling techniques, such as the Data Import Wizard, including Discretization, are presented. Furthermore, we describe a number of information-theoretic measures that will subsequently be required for machine learning and network analysis.
- Chapter 9 deals with missing values, which are typically not of principal research interest but do adversely affect most studies. BayesiaLab leverages the conceptual advantages of machine learning and Bayesian networks for reliably imputing missing values.
- Chapter 12 illustrates the complexity of the seemingly straightforward concepts of attribution and contribution. Bayesian networks can make their meaning clear and their computation possible.
- BayesiaLab methods, functions, and components are capitalized and shown in bold type, e.g., Data Import Wizard. This helps distinguish between natural language expressions, such as "parameter estimation" in general and the Parameter Estimation function in BayesiaLab in particular. Furthermore, concepts that are central to BayesiaLab, such as Entropy and Mutual Information, are emphasized in the same fashion, even though they are not exclusive to the BayesiaLab software.
- User interactions with menus, e.g., the Main Menu or any of the Context Menus, are highlighted with a gray background:
Main Menu > Learning > Supervised Learning > Markov Blanket.
- Keyboard shortcuts are marked with a gray background, e.g.,