• Contribute
• What's New
• Submit News

8.0

Author: Timothy Masters Excerpt: Designed for those with no previous knowledge of neural networks, Practical Neural Network Recipes in C++ serves as a cookbook for neural network solution to practical problems. It will enable those with moderate programming experience to select a neural network model appropriate to solving a particular problem and to produce a work C++ program. The book provides guidance along the entire problem-solving path, including designing the traing set, preprocessing variables, training and validating the network, and evaluated its performance. .: Buy at Amazon.com :. .: Buy at Amazon.co.uk :.

With a title like Practical Neural Network Recipes in C++, you'd expect a 'problem/solution' setup like The Perl Cookbook. With "C++" being in the title, you'd expect C++ classes to be an integral part of the book, most likely with a hierarchy of classes being built up as the book progresses. In fact, none of this is in the book - the code could easily be converted to C, and there is nothing that could constitute a "recipe"! Despite this, the book is still very satisfying.

The best part about this book is how is covers areas of neural networking that should be covered, but are very rarely looked at. Chapter like "Designing Feedforward Network Architectures", "Interpreting Weights: How Does This Thing Work?", "Designing the Training Set", "Preparing Input Data" or "Evaluating Performance of Neural Networks" really tell what you should know about NNs, not just what is cool about them. The depth to which Masters goes to explain these incredibly important concepts is brilliant. I definitely think that these chapters have given me more knowledge about how to apply an NN to a problem than any other literature I've read.

Despite the writing being excellent, both in style and content - I can't really say the same thing about the code. Some of the code was pretty obfuscated - take this example:

*x++ = *center++ + temp * r;

Ok, I'm done bashing the code. I must emphasize that the code didn't let the book down, because the content was so good. Masters manages to present the topics at hand in an excellent way, making it presentable to all. He covers mathematical aspects very well - there are a few equations and mathematical explanations, but nothing that would scare anyone away. For people that do want a more mathematical explanation, he always gives good references for people to look at elsewhere. The book uses diagrams and visual aids very well to help present mathematical formulas or concepts easier as well (the 3D representation of an XOR net output was cool).

All in all, the book was excellent (albeit mis-titled), covering every topic you need to develop and apply neural networks to any problem. For me, though, the code provided was not useful in any way. Luckily, Masters does not rely on the code in any shape or form so it does not affect the overall usefulness of the book.