The Model Development Process

Abstract

In the previous chapter, you saw how an algorithm works. In this chapter, I will review how an algorithm is developed; this obviously is hugely helpful in understanding the many ways biases can creep into algorithms. Also, seasoned data scientists may want to briefly glance at this chapter so that they are aware of my mental frame and terminology since I will be referencing both frequently going forward. One note on terminology: with the advent of machine learning, a whole new vocabulary has been introduced (e.g., observations have become instances, dependent variables have become labels, and predictive variables have become features), which unfortunately makes it really hard to write something that all generations of data scientists can understand. At least the new job title of data scientist is a lot fancier than model developer or modeler, which is what data scientists used to be called in ancient times (ca. anno 2010)! Apart from the title, I will generally use more traditional terms, mostly for the benefit of those who may have had just a tiny bit of exposure to statistics in other fields of study and for whom it will be easier to connect the dots if I use familiar terms.