Big Data/Machine Learning/AI
Epidemiology in space! Reproduce your research anywhere Malcolm Barrett* Malcolm Barrett Stanford University
Voyager I and II, launched in 1977, each have about 70 kilobytes of memory, less than most of the smallest files on our phones. Although 50 years old, scientists that maintain these probes sometimes make impressive code maneuvers to keep the probes communicating with Earth. Yet, here on Earth, we often have trouble running analysis code from last week. Why is it so hard to reproduce code, and how can we make ours more resilient to changing technology? At the Health Policy Data Science Lab, we have cultivated a framework of thinking about reproducibility, durability and duration of code. We provide guidance to our scientists on what techniques to apply, including reproducible documents, version control, package management, and containerization, as well as when to apply them. Our comprehensive approach to reproducibility provides both a way of thinking about reproducibility and practical applications that has been applied to papers in population health, statistical methods, decision science, and more. In this talk, I’ll share our framework and techniques as well as guidance on getting started in your own lab.