This book fills a gap in machine learning literature: all important topics are directly accessible, while at the same time all content is presented mathematically rigorously from a unified perspective.
It's not such a satisfactory read as some more expensive books. However, it offers a couple of insights that are not found in more standard recommendations. There are many unusual approaches such as referencing ROC curves a lot. Also the referenced material and reading suggestions are quite good. The layout is pretty neat too.