"Programming Collective Intelligence" by Toby Segaran. It presupposes no knowledge of statistics or machine learning and focuses on 1) what machine learning technique is probably best in a given situation and 2) how to implement it in Python.

My only complaint is it does completely skip over theory. That makes it accessible to a wider audience (I have a degree in literature) but he misses the opportunity of helping out people who are trying to build a better theoretical foundation or are learning higher math / mathematical notation.