COMPUTATIONAL SOCIAL SCIENCE SEMINAR – Operationalizing Agent Based Models: Introducing Mesa Packages – Tom Pike
COMPUTATIONAL SOCIAL SCIENCE FRIDAY SEMINAR
Tom Pike, CSS PhD Student
George Mason University
Title: Operationalizing Agent Based Models: Introducing Mesa Packages
Friday, April 6, 3:00PM
Center for Social Complexity Suite
3rd Floor, Research Hall
Packages of code are ubiquitous across Object Oriented Programming. Few, if anyone, writing an application programs everything from scratch, instead they rely on optimized packages. Yet, for Agent Based Models the emergent behavior observed in articles or even in class (e.g. CSS 600) is not integrating optimized agent cognition packages or opinion diffusion packages into their Netlogo, MASON, MESA or other ABM platform instead it is using the platform and then coding their ABM from scratch or “hacking” an existing” model. This begs the question, could ABMs also have optimized packages of various agent dynamics so modelers could more easily alter data sources, compare the effects of different agent cognitions or conduct any other myriad of combinations to develop more complex models?
A repository of optimized packages compatible with existing ABM platforms would help catalyze different uses of ABMs. First, it would make it easier to integrate ABMs into policy discussions and policy development as individuals could more easily bring in complex agent behavior to their specific problem. In one sense, this approach creates technical bridge between practitioners and researchers as observed in machine learning and artificial intelligence. Why not the same for the social sciences and policy development? Second, it would help researchers who could now compare different combinations of accepted theories against real world data. Or, allow researchers to focus on one aspect of a complex system as they no longer must code the rest of the system dynamics themselves.
In support of this objective, this presentation will introduce Mesa Packages a distributed repository designed to help share packages for Mesa and discuss one package based on coalition game theory to provide a specific example of how Mesa Packages can work.