Practical Machine Learning: A New Look at Anomaly Detection 1st Edition

I came to the author and book by a personal recommendation and found, like the other review suggested, it's pretty light-weight. Light weight enough that you can do as well, or better, surfing the internet for this stuff. A book should spare you the work of finding and evaluating sources. I didn't connect well enough with this book to think it did. At least i rented the book.Many times I get some better mileage out of either reading the first chapter or two in a more advanced book, or doing that and give a light read to later chapters. The one place this book gets a little unique and interesting is with respect to anomaly detection. I expected a stronger tie in to either computer network intrusion, or how to find ops issues. The EKG example was a little to far from what would be useful at work because the regular or non-anomalous patters weren't that measured or predictable.The author came highly recommended. It's a shame he hasn't written (at least here) to a different audience, as suggested by his response to the other review.

There are a lot of short, introductory texts and review articles out there that are really useful- they introduce you to the fundamental concepts of the field, so that you have a basic understanding and so that you'll know what to look up if you need it. This is not one of those books.The depth of the "practical machine learning" advice in this book is at the level of gems like "before you can spot an anomaly, you first have to figure out what 'normal' is." (chapter 2) Really? My anomaly detection system will have to know what things AREN'T anomalies? Well thank God I dropped $18 to find that out.Sure, the book (sort of) introduces some important concepts that could point you toward more information- like self-information, maximum entropy distributions, type I and II errors, and Bayes risk. I say "sort of" because they're not derived, motivated, or explained in any detail. Most importantly, the authors don't use the proper terms for any of them, so you won't even know what to look up for more information.My favorite chapter is the one devoted to the "t-Digest" algorithm, which was developed by one of the authors. You get to spend the entire chapter waiting for the part where they explain the algorithm, what it does, or how it works. Guess what- it's not there! There's literally an entire chapter on an algorithm that never discusses, even qualitatively, what the algorithm is.I honestly have no idea who this book is supposed to be for. The authors bring up Mahout constantly, which you're probably not using if you're new to machine learning. If you aren't a complete novice, though, you'll just be insulted. And if you have any expertise at all in machine learning or probabilistic modeling, and thought that this book might contain some practical advice for designing anomaly detection systems, you'll be sorely disappointed.Amazon lists this book as being 66 pages, which is only technically true if you count the title page, table of contents, Strata advertisement at the end, and (I'm not making this up) two blank pages. It's a small book with large print, padded with lots and lots of white space and irrelevant photos (like someone holding a magnifying glass over the word "anomaly" on a laptop screen). At some point, apparently, quality control at O'Reilly really went downhill.