Jul 24, 2013 — Archetypal Analysis — Archetypal Analysis is an unsupervised learning algorithm that postulates that each of our set of observations is some combination of some number, K, “pure subtypes,” or archetypes.  I’ll also refer to these as prototypes.  In the simple case of adult human... Read more

Jul 24, 2013 — Non-Negative Matrix Factorization — Non-negative matrix factorization (nnmf) is an unsupervised machine learning method for (you predicted it, strictly positive) data!  The goal of nnmf, of course, is dimensionality reduction.  We start with a dataset, X (size N x p), and want to decompose... Read more

Jul 24, 2013 — Sparse K-Means Clustering "Sparcl" — Tibshirani and Witten introduced a variation of K-Means clustering called “Sparcl.”  What gives it this name?  It is a method of sparse clustering that clusters with an adaptively chosen set of features, by way of the lasso penalty.  This method... Read more

Jul 21, 2013 — Why have clinical decision support systems (CDSS) not worked? — This is by no means an all encompassing list, but I wanted to have a place to jot down some ideas for why CDSS are largely just mediocre.  The big picture summary is that there are problems with the models... Read more