My last review for Manning Publications Co. can’t have been all bad, because they asked me to do another one. The book this time is Machine Learning in Action by Peter Harrington and currently available in MEAP. The review is a so called “2/3” review as the book is not complete. I had the pleasure of reading chapters 1 through 10 as well as some of the appendixes.
Many years ago I did my Master thesis on the mathematical properties of artificial neural networks. I haven’t been working much in this area lately, so it was quite fun to “re-visit” some of the theory and applications used both in ANN and machine learning.
The book is very application oriented and gives some very good and illustrative examples of algorithms which can be used for classification, forecasting or unsupervised learning. It uses Python for all the code examples, but gives very good directions on how to install and use it, so if you are new to Python or have never used it before, this should not hinder you from getting value from the examples.
The author has included a lot of references to background material, enabling the reader to seek more information on areas of specific interest.
One area I find is a little weak is the handling or explanation of the underlying statistics. Machine learning is really just a form of Non-linear optimization and we know when these models are better then OLS (Ordinary Least Squares) or “regular” regression. If a certain set of conditions are met, the OLS estimate is the Maximum Likelihood (ML) estimate, and then we really can’t do any better. What this means is, that machine learning or neural networks or whatever we call it, will only be better if these conditions are not met.
One could fear, that the inexperienced user, would draw conclusions which on first sight would seem correct, but which would actually be wrong, because the underlying model was incorrect or the supplied data did not support it.
This being said, I found the book a very good read and a good introduction to Machine Learning.