25 of 25 people found the following review helpful:
5.0 out of 5 stars
Very practical indeed, August 5, 2001
This review is from: Practical Neural Network Recipies in C++ (Paperback)
This book is exactly as advertised. Other excellent books on Neural Networks will have you buried in mathematical notation that will challenge even readers with some statistics background and a couple of semesters of Calculus. These books are definitely worth your while if you can handle the math, but even then, translating these books from theories to solving real problems is no easy feat. By contrast, this book presents a good introduction to basic feedforward neural networks that is very readable to users with a moderate math background, and probably readable with some effort for motivated readers with limited math. You can read this book and come away with a reasonable understanding of how a feedforward network functions. Still, that's not even the strength of this book.
Not only is this book "practical" in the sense that it is readable, it is practical in that it tackles a host of additional topics necessary for using a neural network in the real world. It discusses annealing and genetic techniques for avoiding local minima. It discusses singular value decomposition for avoiding problems with redundant inputs. It discusses the best ways of building training sets and preparing input data, as well as ways of evaluating the performance of networks and attaching confidence measures. It would be easy to charge right in, use a neural network as a black box, give it a dataset and train it, and then wait for it to pop answers out. The only problem is, this will yield results that are worthless in the real world. All of these concerns have to be addressed to build a model that can actually be used for something.
I was very happy with the code base included with this book as well. In addition to a neural network using conjugate gradient descent (as well as Kohonen learning), code is integrated into the main program for annealing, genetic initialization, and singular value decomposition, as discussed in the text. I found the section on how to use the program slightly confusing at first, but once I figured out how to operate it, it was easy to set it up and use it. The code base is C++ that is deeply rooted in C, so it won't impress object-oriented gurus at all, but it should be understandable and fairly easy to work with for users with a good background in C, but who aren't C++ experts. For me, the bottom line is that the code works, it's not hard to understand (in my opinion), and it shouldn't be that hard to extend to perform new functions. In this day and age, it's probably worth mentioning that the program comes with a simple command-line interface, so if you want something that runs in a spiffy GUI, you'll have to write one.
I would recommend this book strongly as a first book on neural networks for readers that are interested in learning neural networks in the context of solving practical problems. I would also recommend this book to readers who have a book or two discussing the theoretical aspects of neural networks and want something that will help them translate that into attacking practical problems, and also provide a code base that will give them a head start.
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
12 of 12 people found the following review helpful:
4.0 out of 5 stars
A good start, May 23, 2000
This review is from: Practical Neural Network Recipies in C++ (Paperback)
Some of the other reviewers of this book must have suffered from a misconception about the book. It is exactly as advertised, if you don't think so, compare it to Neural Networks, A Comprehensive Foundation by Haykin or Artificial Neural Networks by Schalkoff. Those are REALLY academic. Neural Networks is a very difficult topic,but this book does the best job I've seen yet of explaining Neural Nets in a Straightfoward, understandable way. C++ Neural Networks & Fuzzy Logic by V. and H. Rao tried this and failed. The math is very needed, and I respect the approach of only looking at one type of neural net (feedfoward 3 layer) in depth rather than a billion short, unexplained looks a many. Yes, the code is not the best I've ever seen, and it gets a bit rough to follow, but it explains the ideas. Overall I'd say know a little about what you're getting into before buying ANY book on Neural Networks.
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
10 of 10 people found the following review helpful:
5.0 out of 5 stars
How to build 'em - How to use 'em - And actual source code!, February 13, 1997
By A Customer
This review is from: Practical Neural Network Recipies in C++ (Paperback)
Easily the best treatment of neural networks I have ever read. Outstanding treatment of the innards, how they work, and years of practical experience boiled down into heuristics for programming (with optimized source code examples!), configuring, training, and evaluating nets. The theory is brilliantly explained within each topical context in lieu of boring chapters on NN theory and math. Mathematical expressions are used only where they add clarity and are not gratuitiously used where the author's excellent English can do the job. And talk about English! Masters is one of those phenoms who speak math and English with equal facility. The writing is simply outstanding. The book is so good it is hard to decide what parts are most valuable. Amazingly, it is as useful for the novice wanting to learn something about neural nets as it is for a professional looking for tips and techniques! I have made the book mandatory reading for my team of knowledge discovery scientists and engineers
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
