Top positive review
5.0 out of 5 starsThis book isn't perfect, but I'm still giving it 5 stars because ...
October 18, 2015
This book isn't perfect, but I'm still giving it 5 stars because it provides a better overview of the entire field of Machine Learning than any other book I've come across. My background is in computer science and software engineering and I've been interested ML as more of a hobbyist and outside observer for a few years (reading some books, taking Andrew NG's coursera course), just recently dabbling in some applications professionally. What I was still missing before reading this book was as high a level understanding of where all of the models and technique in the field of ML fit. Other books describe the difference between supervised and unsupervised learning, but this book goes further in describing how, say, decisions trees, support vector machines and deep neural networks fit compared to each other and within which subfields statistics play a larger role than others.
The book also puts many techniques in historical perspective that I found very helpful, such as the rise, fall and rise again of deep neural networks with support vector machines taking a lead as the hottest technique in between (while also making clear that SVMs are a useful technique with unique strengths today). Finally, it makes clear that these techniques are not all competing for being the best overall at everything, but that they can be used quite complementary and/or they have unique strengths within certain problem domains. The book accomplishes all of this through a survey of broad subfields of ML, how each has attempted to be *the* master algorithm, has fallen short in some ways, but remains the best at some things and could play a role in the state of the art master algorithm (while acknowledging we're not quite there yet). So while the term 'master algorithm' is somewhat of a gimmick (as he acknowledges), it's a good way to think about what ML is attempting to accomplish as a field: building working, adaptive software systems with less and less human assistance by learning from data, and to see how many specific techniques have played a role in progress.
What I don't know is how accessible this book might be to someone who's less technical. I think the first couple chapters would be a great read for anyone with a general interest, making clear how ML differs from the traditional software / automation that has brought us so far, but it could be that the details within the rest of the chapters that go into more depth would be too in the weeds.
I've also read some other reviews from technical readers that assert the book lacks enough depth to be helpful, but this wasn't the case for me, in fact the level of detail was perfect—just deep enough to match with details I'd skimmed before in previous surveys of the field yet not so deep that I couldn't get through and enjoy the chapters in a casual evening read. The author also explained some concepts better than I've read anywhere else before, such as the debate is between frequentist and bayesian statisticians.
