My journey from R to Julia

My history with R

For 15 years I taught “Applied Epidemiology using R” at the UC Berkeley School of Public Health. I started teaching this course in the early 2000s when most people were not interested in R. Eventually R’s popularity grew, and so did my course enrollment. I taught basic R programming applied to epidemiologic concepts. Students worked on a project of their choosing. Because I had students from multiple disciplines, their projects were often very innovative and I learned a lot from them.

Over the 15 years I witnessed the emergence of “data science” and students’ ingenuity and creativeness in their projects. Naturally, the course evolved too. I introduced Bayesian networks as a unifying framework to introduce probablistic dependence, causal graphs, and decision networks (for decision analysis).

I started dabbling in Python. Unfortunately, the COVID-19 pandemic interrupted my teaching because I was the health officer of San Francisco and director of the Population Health Division. Working on the pandemic response left me little time to conduct analyses or to learn Python.1

I cannot remember when, but eventually I discovered Julia—a programming language designed for scientific computing with the intuition of Python or R, but with the speed of C++. I fell in love with Julia and I gave up on learning Python. I did not, and do not, have the time to maintain core competency in more than 1.5 programming languages.2 As I learned more Julia, I became convinced that for me learning Julia was a better long term investment than sticking with R.

I use programming to explore or test my epidemiologic intuition, to learn new methods, and to visualize and analyze data. I have a personal interest in Bayesian networks, decision networks, causal inference, Markov decision processes, and agent-based modeling.3

Here are some key Julia features that I believe epidemiologists will value:

• multiple dispatch
• composite types
• just-in-time compiling
• speed (very fast)
• community

Julia enables more intuitive programming. For example, in R, we try to avoid loops because they are very inefficient. In Julia, loops are efficient because it compiles before execution. This promotes programming that is more natural.

Below I demonstrate multiple dispatch with a trivial example. R is single dispatch.