Abstract

Modelling epidemics is challenging: realistic models require practitioners to be well-versed in epidemiology, and computational statistics, and computer science. The result is that modelling is in the hands of the few; that models are slow to code and their results opaque; and that there are substantial skills barriers in training newcomers to epidemiology. In this talk, I will give an overview of how programming languages research can help with these challenges. Domain-specific languages are programming languages targeted at a particular problem domain, and allow developers to write succinct and targeted code that can be compiled in a performant way. There have been several promising pilot DSLs targeting infectious disease modelling, particularly those based on reactive programming. We hypothesise that main missing piece of the jigsaw is support for probabilistic programming functionality, especially support for model parameter inference and the ability to deal with censored data. I will describe early thoughts on a proposed system, Amethyst, that aims to serve as a new domain-specific language for Bayesian infectious disease models, using a declarative and compositional approach to support rapid and reliable model development .