Towards Transparent Agent-based Models: Visualizing ABM Process Flows from Simulation Traces
Résumé
One of the challenges in promoting the Agent-based Modeling approach and communicating this kind of model is making it
transparent to stakeholders. Through the years there has been an observed fast development of Agent-Based Models (ABM),
especially in modeling social-ecological systems, which quickly raises and heightens the need for a transparent view of these
models. Various protocols have been proposed for documenting these models, and calls have been made to ensure open
access to the simulation code. However, the lack of consistency in model documentation plus the additional layer required
to streamline comprehension and save time in understanding model code does not fully promote a sufficiently transparent
view for general users. In response, we present a three-part methodology to make agent-based models transparent. Our
objective is to provide a visualization of the ABM processes in a way that closely aligns with the source code without
deriving the information from the model documentation. The approach is as follows: (1) automatically store execution logs,
(2) apply process mining techniques to automatically discover agent processes, and (3) visualize the processes using graph modeling. Simulation logging was implemented within the GAMA platform to ensure compatibility with other models. For
graph modeling, we present the final visualization using Business Process Modeling Notation (BPMN), a standard graphical representation for specifying business processes. The resulting visualization is designed to complement the model’s source
code and serve as an automatically generated model description. We demonstrate our methodology using various toy models
available in GAMA.
Origine | Fichiers produits par l'(les) auteur(s) |
---|