A review by bob_muller
Bayesian Reasoning and Machine Learning by David Barber

2.0

Rant on. I'm afraid I found this book largely unusable as a reference for machine learning. It is possibly good as a high-level introduction to the mathematics of certain kinds of machine learning (note SVM isn't discussed, it "doesn't fit well with the Bayesian approach"). I didn't find the math particularly well explained, and that math assumes far more math and prob/stat exposure than the author claims in the front of the book (upper-level undergraduate with just linear algebra, but the book is filled with partial differentiation and tons of assumptions about probability, Bayesian methods and practices, and math modeling conventions). He almost completely ignores the practical aspects of Bayesian analysis (MCMC and so on) in favor of the very old way of doing things with "conjugate" distributions and so on; nobody does things this way anymore, at least not in the machine learning circles I run in. His approach may be a bit too dependent on the EM algorithm, I can't really judge that. The examples are cursory and usually don't include actual calculations or steps showing how things are actually done in practice. There are no answers to exercises. The code is spotty at best and is done in Matlab, placing it solidly in the "academic" machine learning framework rather than a more practical place. Add to that the fact that I got a copy of the book that had been misbound with missing pages, which I had to return for a replacement. Quality issues at all levels, then. Oh well, rant off.