Customer Reviews

Bayesian Logical Data Analysis for the Physical Sciences: A Comparative Approach with Mathematica® Support

5 star:		(3)
4 star:		(0)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

Phil Gregory has managed to condense the most important aspects of Bayesian probability and data analysis into a book that is actually rather practical. This book will give you a solid Bayesian understanding of probability, starting with first principles (Cox's desiderata), continuing to Bayes theorem, an introduction to common probability distributions, and concluding with rather advanced numerical techniques such as tempered Markov Chain Monte Carlo. The book is geared towards a reader who will use Mathematica to work through examples, but can be successfully used by others who prefer cheaper and more practical computational frameworks. It's not a flawless book by any means---first of all, although the book purports to cover frequentist alternatives to Bayesian methods, the frequentist coverage is very shallow and inadequate to give the reader enough background to either use or really understand frequentist usage. The chapter on maximum entropy techniques is woefully incomplete, and doesn't include general Jeffreys priors (derived from the Fisher information) or really explain the various issues associated with defining the entropy for continuous distributions. The section on deriving priors with uncertain constraints actually doesn't give an answer to how to handle uncertain constraints! But on the plus side this book answered a number of questions that have long puzzled me, such as why frequentists marginalize over nuisance parameters by minimizing the likelihood function with respect to them instead of integrating over them (it's a dodge that only really works for Gaussian-like distributions), and how to handle the enormous numbers of parameters that Bayesian calculations can generate through the use of Markov Chain Monte Carlos. I wish the book had been longer or more detailed, but if you want to learn Bayesian analysis and don't care much about understanding frequentist statistics, this is an excellent place to start.

Nicely written. Clear and concise. Chapter 12 contains a great introduction to Monte Carlo methods, the best I ever had read for novice practitioners. Buy it!

Very nice book. Even if you don't use Mathematica (I only use it on occasion), this is a very insightful book on Bayesian inference. The future is Bayesian, this book will help guide you to that future. Yes, there are a lot of other books on Bayesian inference on the market, but I really think this has something to offer and is complementary to many of the other books. Being a physicist, Gregory pays attention to Jaynes' brilliant work (especially since he was introduced to it when he found the PhD dissertation by Jaynes' student, Bretthorst, in a restroom one day!), which is light years ahead of its time. If you want to be an ace data analyst/statistician, this book will give you put you in with the front runners.