Review: Bayesian Statistics the Fun Way by Will Kurt
The main conference hall enveloped me in its gentle, air-conditioned embrace. Outside, in typical New York City fashion, the sun burned and the heavy, humid air threatened to suffocate everyone indiscriminately.
After registering and picking up my badge, I found myself drawn immediately to one specific vendor booth. This vendor was No Starch Press, a publisher of fine works centered on software.
I have a weakness for purchasing books. It’s only gotten worse with time, so I built up a habit of restraint that kicks in, automatically, whenever I find myself where the printed word is transacted. It’s 98% effective. But this time, I walked away, grinning, with a crisp, slim book called “Bayesian Statistics the Fun Way” by Will Kurt.
Kurt begins by describing what probability is. Then he proceeds to build a platform made up of the basics: fundamental notation & operations; the binomial, normal, and beta distributions; conditional probability; and Bayes’s theorem. Onto this platform he hoists more complex topics like parameter estimation, Bayes’s Factor, and hypothesis testing. All of that in under 200 pages (the rest are appendices), which the author suggests could be covered over a longer flight.
That’s possible. Kurt’s style is smooth and light, which allows for quick progress. More importantly, though, the knowledge is laid out so that each chapter feels like just a small inferential step away. Occasionally, a chapter backtracks in order to offer a new perspective on an earlier topic. All of this combines into an electrifying effect; almost every page yields a new, satisfying insight.
I can’t help but remind myself how the same material was treated in high school. There the knowledge was spread out in a fifth of the space—and most of that was devoted to lists of problems. (Great for homework and testing, probably). Perhaps it was my teenaged brain, but absolutely nothing connected: not permutations, not distributions, not Bayes’s theorem. None of it also had any relation to reality whatsoever—what madman carries on their person loaded dice or bags upon bags of many-colored marbles?
But I digress.
Of particular help were the few problems that closed off each chapter. At first, they were simple enough to solve with pen and paper, but later I had to bring out my laptop and fire up Jupyter Notebooks. (The book presents the material using R, but I had no difficulty adapting everything to Python+numpy+scipy). Their value lay in making me immediately aware that I had missed some important idea or two, prompting me to go back and extract yet more meaningful insights, further fanning my curiosity.
I would be remiss to not mention how heavily the author leans on using examples. For a mathematical rookie like me, this was welcome help—it turned abstract concepts into intuitive, almost tactile ones. In one instance, Kurt uses lego bricks to visually show how different parts of Bayes’s theorem come together, turning what had until then been a memorized formula into a working conceptual tool. (Why isn’t this mode used more often by educators?)
Today, when the wind carries the icy smell of winter, I know it wasn’t luck that put this book in my hands. Rather, it was high likelihood, what with being at a hacker conference, combined with a high prior—the result of years spent lurking the rationality web.
If you feel curiosity toward statistics and probability, Bayesian Statistics the Fun Way is a great book to get started with. Not only does it provide a solid, practical foundation, it also delivers on the promise from the title: it really is fun—that’s my excuse for diving into Markov Chains! So be careful, you might get pulled in deeper than you thought you would!