Books to Read

Here are some interesting books that I have been reading and would recommend to others.  Some of the books are general in nature and accessible to a wide audience, others can be quite technical.

The Undoing Project, Michael Lewis (General):

This is one of the best books that I have read in years.  It is the story of  two Israeli psychologists Daniel Kahneman and Amos Tversky who have made their mark on and to an extent created the science of behavioral economics.  Kahneman eventually went on to win the Noble prize in economics, and last year’s winner, Richard Thaler, was also heavily influenced by them.  The book is part biography, and part intellectual adventure into the theories of these two researchers. The unusual relationship between the authors is described, with a focus on how their theories thrived from their collaboration. Throughout the book, the theories of Kahneman and Tversky are described and the unusual way in which we make decisions is described.

Burn the Business Plan, Carl Schramm (General)

I worked with Carl many years ago when we sat on the board of a healthcare analytics company.  Carl has an exceptionally keen intellect and is a leader in education of entrepreneurship, having led the Kauffman Foundation for many years.  He dispels the myths that you need to be a young (male), high tech wizard to start a new company and that every new company must start with a lengthy, detailed business plan.  He in fact suggests that business plans motivate for a build and flip ideology, which is in fact counterproductive for building a solid business and at odds with the way that most successful companies were in fact started.

Bayesian Regression Modeling with INLA (Technical)

Although the Bayesian approach was developed about 250 years ago, it is really just in the last 20 to 30 years that the computational power and methodologies have been developed to leverage the flexibility and power of the approach.  A key insight is that Markov Chain Monte Carlo (MCMC) techniques can be used to “solve” arbitrarily complex Bayesian models.  Yet even today, MCMC tends to be slow.  For very large data or highly complex models, MCMC may be too slow to be practical.  INLA is a software package that uses an approximate technique, Laplacian Approximation, to vastly speed up computation and it produces accurate results for a wide class of models.  This book offers the first step-by-step guide to using INLA.

Over the course of the next few months, I will be discussing various different Bayesian models and techniques in future blog posts, including the use of INLA.

Free Technical Internet Books

There are a number of machine learning / statistical books that are freely available on the internet and have remarkably good content.  Here are a few of them:

  • Gaussian Processes for Machine Learning, Rasmussen: Gaussian Processes (GP) are a remarkably powerful and flexible means of modeling non-linear data.  They are particularly useful in areas where there is limited data and understanding the error  rate in modeling, as opposed to the best point estimate, is important.  This book is the seminal book on GPs and although old is still the key resource for modeling with GPs.
  • Advanced R, Wickham:  The top two software platforms for doing machine learning and statistical analysis now appear to be Python and R.  Each have their relative advantages and disadvantages.  This book is an exceptionally good resource into some of the more advanced topics in R.
  • Elements of Statistical Learning, Hastie et al.  This is the classic book on machine learning, and while dated, still should be a staple in every library for those doing machine learning.