Denis, Jean-Baptiste_ Scutari, Marco-Bayesian Networks With Examples in R-CRC Press Taylor & Francis Group (2014).pdf
Applications of Bayesian networks have multiplied in recent years, spanning
such different topics as systems biology, economics, social sciences and medical
informatics. Different aspects and properties of this class of models are crucial
in each field: the possibility of learning causal effects from observational data
in social sciences, where collecting experimental data is often not possible; the
intuitive graphical representation, which provides a qualitative understanding
of pathways in biological sciences; the ability to construct complex hierarchical
models for phenomena that involve many interrelated components, using
the most appropriate probability distribution for each of them. However, all
these capabilities are built on the solid foundations provided by a small set
of core definitions and properties, on which we will focus for most of the
book. Handling high-dimensional data and missing values, the fine details of
causal reasoning, learning under sets of additional assumptions specific to a
particular field, and other advanced topics are beyond the scope of this book.
They are thoroughly explored in monographs such as Nagarajan et al. (2013),
Pourret et al. (2008) and Pearl (2009).
The choice of the R language is motivated, likewise, by its increasing popularity
across different disciplines. Its main shortcoming is that R only provides
a command-line interface, which comes with a fairly steep learning curve and is
intimidating to practitioners of disciplines in which computer programming is
not a core topic. However, once mastered, R provides a very versatile environment
for both data analysis and the prototyping of new statistical methods.
The availability of several contributed packages covering various aspects of
Bayesian networks means that the reader can explore the contents of this
book without reimplementing standard approaches from literature. Among
these packages, we focus mainly on bnlearn (written by the first author, at
version 3.5 at the time of this writing) to allow the reader to concentrate on
studying Bayesian networks without having to first figure out the peculiarities
of each package. A much better treatment of their capabilities is provided in
Højsgaard et al. (2012) and in the respective documentation resources, such
as vignettes and reference papers.