The code is the model
- University of Zurich, Switzerland
Figures
A Git client.
This screenshot shows how the Git client SourceTree lists changes Steve Phelps made to the Java Agent-Based Modeling library (github.com/phelps-sg/jabm), which was used to produce the results presented in Caiani et al. (2014) and Caiani et al. (2015). The selected commit to the yellow branch with fingerprint c282914 apparently fixed the size counter of a list of agents named AgentList. When not only commenting on technical decisions, but also modeling considerations, Git can serve as electronic lab journal that is tightly linked to the relevant source code.
Waterfall modeling.
Notes: This diagram depicts the waterfall view of the modeling process of a computer simulation and its accompanying paper. It is highly idealized and hard to follow in practice. There is no indication on how to handle errors or how to otherwise refine the initial idea. Yet, this is often how the development process is implicitly assumed to work.
Agile modeling.
Notes: This diagram presents a more realistic view of the modeling process, incorporating tight feedback loops to decrease the costs of testing ideas. Formal specifications got replaced with a set of temporary documents that help communicating and refining ideas. Unlike before, ideas can also be formulated directly in a programming language. From there, three feedback loops around different classes of errors help to decrease the time it takes to ensure the code works as intended. Once it does, the simulation outcome is used to refine the initial idea, closing the outer, bold feedback loop. Each time it is completed, the modeler learns something about her model and assumptions.